Framework of Students’ Learning Pathway in an E-Learning Setting Integrated With Learning Analytics Intervention Based on Learning Styles and Preferences of Students

Abstract

Learning Analytics intervention is an important approach that helps teachers and students improve the learning process and performance, especially for those who need more support. However, there has been limited research on developing Learning Analytics interventions based on students’ learning styles in e-learning, aiming to enhance key aspects like motivation, engagement, retention, and academic achievement. There are also fewer studies on understanding students’ learning pathways. This research designed a new Learning Analytics intervention in an e-learning environment. It examined students’ learning performances and formulated a framework of students’ learning pathways in the e-learning environment integrated with the Learning Analytics intervention based on their learning styles and preferences. An experimental design was adopted with a population of Year Two undergraduate students, employing several validated instruments. The collected data were then analyzed using descriptive analysis, content analysis, and data mining. Results indicate that most students’ learning performances were enhanced, and students exhibited different behaviors in terms of the number of log-ins, views, and posts created. Utilizing these findings, a framework of students’ learning pathways in the e-learning environment embedded with the developed Learning Analytics intervention was successfully formulated using WEKA data mining software. While the Learning Analytics intervention was designed to accommodate multiple learning styles, the framework development and detailed analysis were focused on visual learners due to the predominance of this learning style in the participant group. This framework may serve as a valuable reference point for future research.

Introduction

With the presence of technology and big data in the educational field, universities are increasingly equipped with different technological facilities and learning support services. One example is the application of e-learning, which has boosted new teaching and learning experiences for educators and students (Derbas, et al., 2025; Rodrigues et al., 2019), as it gives students more options and empowers them to assume accountability for their own learning (Y. Yang et al., 2025). Nonetheless, students sometimes face challenges when learning through online courses (Kew & Tasir, 2022a), such as lacking motivation (Esra & Sevilen, 2021), struggling with academic performance (Chiner et al., 2021), low cognitive engagement (Johar et al., 2023), and difficulty retaining information (Bi et al., 2023). Previous research has emphasized the need to create effective e-learning environments and learning materials that cater to the diverse learning styles of all students (El-Sabagh, 2021; Rana & Lal, 2014). Additionally, studies have suggested providing targeted interventions according to students’ individual learning preferences to improve learning standards in education (Nguyen et al., 2024). Hence, effective technology implementation in learning is essential to meet students’ learning-style preferences (Hsu et al., 2023), enhance their learning performance, and solve learning issues in e-learning. Thus, Learning Analytics (LA) is one approach that could be used to address this matter.

As indicated in the literature, the role of LA is crucial in gathering and examining educational data to unveil valuable insights into students’ activity and behavior within e-learning (Heikkinen et al., 2023). Its significance grows as it facilitates a data-driven comprehension of the occurrences in e-learning. As students engage in various online learning activities throughout their academic journey, they generate a plethora of data-rich footprints, encompassing metrics like clicks and posts. Educators can seamlessly download, track, and analyze these digital traces, utilizing the information to understand students better, address challenges inherent in online learning, and offer targeted intervention support when needed. LA can also help understand how students behave in e-learning and what their learning pathways look like, which can give educators a deeper understanding of their students’ behaviors in e-learning, such as their navigation patterns and the challenges they face and provide essential interventions to enhance their learning performance. Thus, Kew and Tasir (2024) verified that the success of LA hinges on its ability to present data to educators in formats that facilitate informed decision-making regarding educational interventions. Similarity, Wong and Li (2020) and Heikkinen et al. (2023) reported that the implementation of LA interventions can enhance students’ learning outcomes.

Nonetheless, research on LA intervention design (e.g., providing students with suitable/preferred learning objects, posting the signal, etc) is still deficient to assist students, particularly those who are at risk (Kew & Tasir, 2017; Wong & Li, 2020; Wu et al., 2015; C. C. Yang & Ogata, 2023). Similarly, Wong and Li (2020) highlighted that the intervention model is a useful yet less studied component in research. In fact, intervention is the most challenging aspect in LA (Rienties et al., 2017; C. C. Yang & Ogata, 2023). Therefore, to use the student-generated learning data and optimize students’ learning, it is essential to examine how LA can be integrated into e-learning environments. This can help align LA’s potential with students’ diverse learning style preferences and address any learning challenges they face. To fill the research gap in LA and enhance students’ learning performance, this study developed an LA intervention embedded within an e-learning environment. The intervention was developed with consideration of students’ individual learning styles, aiming to furnish them with personalized learning materials. The key objectives of the study were: (a) analyze students’ learning performance in the e-learning environment with the LA intervention, focusing on factors like motivation, cognitive engagement, retention, and academic achievement; (b) examine students’ learning behaviors in terms of their login frequency, content views, and posts created within the e-learning environment with the LA intervention. Based on these two findings, it aims to (c) formulate a framework of students’ learning pathway in e-learning integrated with the LA intervention based on students’ learning styles. However, although the Learning Analytics intervention aimed to address various learning styles, the framework’s development and in-depth analysis concentrated on visual learners, as this style was most prevalent among the participants. The study aimed to offer valuable insights into leveraging LA to optimize the e-learning experience and support students’ diverse learning needs.

Literature Review

Learning Issues in E-Learning

Learners often prefer online learning due to its clear structure and efficient information delivery (Pokkuluri et al., 2023). E-learning technology enhances traditional education by improving teaching and learning practices (Sadeghi, 2019). However, issues remain, particularly the mismatch between provided learning materials and students’ diverse learning styles (Rana & Lal, 2014; Kew & Tasir, 2022b). This gap often results from limited understanding of learners’ preferences and personalities in online environments, despite the growth of e-learning in higher education. Many e-learning systems deliver uniform content without considering individual differences (Liu & Yu, 2023). Understanding learning styles can help educators tailor instruction to enhance academic performance (Heikkinen et al., 2023). As Dantas and Cunha (2020) emphasized, incorporating learning styles can make the learning process more effective. Thus, aligning learning materials with students’ preferences and personalities is crucial.

Retention is another major concern in e-learning. High dropout rates continue to affect academic success (Aldowah et al., 2020; Bi et al., 2023), with low motivation being a significant contributing factor. Motivation influences student engagement and is key to learning success. Unfortunately, students’ cognitive engagement in e-learning remains low despite technological advances (Johar et al., 2023; Kew et al., 2018). Engagement is also an important indicator of teaching quality (Johar et al., 2023; Zhang et al., 2019). Therefore, addressing motivational and psychological aspects is essential in designing more effective online learning environments. LA offers a promising solution by enabling targeted interventions to support struggling students, improve engagement, and enhance overall learning quality (Wong & Li, 2020; C. C. Yang & Ogata, 2023).

Learning Analytics Intervention

The role of Learning Analytics (LA) is vital in enhancing education quality by providing evidence of students’ learning patterns and behaviors, including their learning pathways and navigation (Viberg et al., 2018). These insights help educators and learners adjust teaching and learning strategies based on individual student needs, aiming for timely and effective interventions. Studies have shown that LA interventions in e-learning address several challenges, such as increasing student engagement, improving retention, boosting performance outcomes, and enhancing motivation and satisfaction (Kew & Tasir, 2022a; Lonn et al., 2015). LA tools support personalized learning by helping identify students’ needs and tailoring interventions accordingly.

LA allows educators to adapt the learning environment in near-real-time to better suit each student (West et al., 2018). In online learning, it enables the creation of personalized learning materials aligned with individual learning styles. Van Harmelen and Workman (2012) demonstrated how LA can detect at-risk students and apply targeted interventions to keep them engaged. The evolving nature of LA interventions further enhances their effectiveness (Wong & Li, 2020; C. C. Yang & Ogata, 2023), helping educators foster logical thinking and make data-driven decisions. Research also highlights practical applications of LA. For instance, Wise et al. (2014) used LA to support online discussions, while Choi et al. (2018) tracked click data to intervene with students at risk through emails or consultations. Despite these benefits, recent reviews show a lack of interventions tailored to students’ learning styles and the continued reliance on one-size-fits-all approaches (Kew & Tasir, 2017; Wong & Li, 2020). Overall, LA serves as a powerful tool to improve online learning, enabling educators to deliver more personalized and effective educational experiences.

Related Studies: Learning Analytics Framework

Several studies have proposed Learning Analytics (LA) frameworks aimed at improving educational outcomes. Nguyen et al. (2024) introduced the FARCL framework to assess self-, co-, and socially shared regulation in collaborative learning, aligning LA with theoretical foundations. Alalawi et al. (2025) evaluated the SPPA framework, which helps educators identify at-risk students and apply targeted interventions. Their results showed improved pass rates and reduced failure rates. Similarly, Caporarello et al. (2019) proposed an extended LA framework for higher education that centers on learners and instructors to personalize learning paths and enhance teaching effectiveness. Collectively, these frameworks emphasize regulation, prediction, and personalization. However, none explicitly address learning pathways or student engagement based on individual learning styles.

A further search using keywords like “Learning Analytics Intervention Framework” and “Learning Analytics Intervention Model” revealed a limited number of studies, indicating that LA interventions are still in development, summarized in Table A1. The literature suggests a clear need for interventions tailored to learning styles, as these profoundly impact student engagement, comprehension, retention, and academic performance (El-Sabagh, 2021; Tie & Umar, 2010). Gašević et al. (2017) also emphasized aligning LA interventions with students’ needs. However, existing research rarely addresses how learning pathways can be shaped according to learning styles (Lin, 2016). While You (2016) examined learning behaviors influencing performance, these were not tied to learning styles. To address these gaps, the present research aims to develop an LA intervention based on learning style models for e-learning. This includes providing learning objects tailored to individual preferences to enhance motivation, academic achievement, cognitive engagement, and retention – ultimately contributing to more effective, personalized e-learning experiences.

The Design of New Learning Analytics Intervention Based on Learning Styles

The new LA intervention in e-learning was developed by integrating the Felder-Silverman learning styles model (FSLSM) and Keller’s ARCS model. This approach aims to provide personalized learning objects that enhance students’ motivation, academic achievement, cognitive engagement, and cognitive retention. The development process follows the LA Cycle (Clow, 2012, as cited in Kew & Tasir, 2022a), as shown in Figure 1. This research aligned the LA intervention process with Clow’s LA Cycle (i.e., learner, data, metrics, and intervention) to enhance theoretical coherence. In step 1, learners who were taught and asked to complete tasks in e-learning were chosen: the population of 50 Year Two undergraduate students who took a computer-based subject and used e-learning were selected. In step 2, data generated by these students related to their learning behaviors were tracked and collected. This research collected students’ log files, which were retrieved from the Learning Management System platform, Moodle, which assisted in providing data generated by students to identify their learning behaviors in the e-learning environment. Students and instructors were assigned roles through authenticated login via the university’s e-learning system, and Moodle’s built-in log and report system was enabled to monitor student engagement (e.g., number of logins, content views, forum posts).

Figure 1.

Development process of LA intervention.

In step 3, the metrics were carried out. In this study, the researchers analyzed different aspects of the students’ learning experiences in e-learning, such as the students’ learning styles, motivation levels, academic performance, engagement, and retention of information. Various analysis methods, including descriptive analysis, inferential analysis, content analysis, and data mining, were used to analyze the data. Based on the analysis results, the researchers identified which students were considered “at-risk” and which were “not at-risk.” Many researchers have indicated that students achieving grades of B- and below are classified as being at risk (e.g., Bainbridge et al., 2015; Baker et al., 2015; Er, 2012; Hu et al., 2014; Tarimo et al., 2016). Accordingly, this research identified students as being at risk if they scored B- and below (regarded as low) in cognitive retention and academic performance tests. Thus, in this research, motivation level, cognitive engagement level, cognitive retention, and academic performance are the variables used to indicate whether students are at risk. In other words, the at-risk students had lower levels of motivation, academic achievement, cognitive engagement, and cognitive retention (Kew & Tasir, 2022a). During the cleaning data process, the log data extracted from the Moodle LMS was cleaned and prepared in WEKA 3.7.12. Irrelevant attributes (e.g., system IDs, incomplete logs) were removed, and missing values were handled using WEKA’s ReplaceMissingValues filter. Next, to classify the students as at-risk or not at-risk, the researchers used a decision tree technique analysis tool called WEKA 3-7-12. WEKA was used to classify students who were at-risk and not at-risk. The findings showed that 20 students (40%) were not at-risk, while 30 students (60%) were considered to be at-risk. The Kappa statistic is 1, and the ROC area is 1, indicating that the classification is accurate. In other words, it confirms that the way to classify at-risk students based on academic achievement is correct. After this categorization, the researchers then looked at the students’ learning styles. This was done so that the appropriate learning assistance interventions could be provided to the at-risk students according to their individual learning styles. This step is crucial. The at-risk students exhibited diverse learning styles, including intuitive, sequential, active, global, verbal, and visual. This means the LA intervention in Step 4 should incorporate more learning objects that cater to these varied preferences, along with elements of the motivational model for the entire class.

In Step 4, the LA intervention was developed and embedded within the e-learning environment. It integrated the Felder-Silverman Learning Style Model (FSLSM) and Keller’s ARCS motivational model. These two models were crucial in enhancing students’ motivation and addressing the needs of learners with different preferences. The specific motivational strategies from the ARCS model that were used to develop the LA intervention are presented in Table 1. Regarding attention, questions are used to attract students’ attention, while set goals related to students’ experiences are used in the relevance category. For the confidence category, students are given instructions to make them confident about achieving success. Meanwhile, the satisfaction category provides positive reinforcement to encourage students.

Table 1.

Keller’s ARCS Model.

Category	Basic strategies
Attention	Inquiry arousal: stimulating an attitude of inquiry
Relevance	Goal orientation: meeting students’ needs
Confidence	Learning requirements: building a positive expectation of success; Success opportunities: increasing students’ beliefs in their competence
Satisfaction	Positive results: providing reinforcement to students’ success

Source. Adapted from J. Keller and Suzuki (2004).

The Flexible Systematic Learning Style Model (FSLSM) is a well-known model used in online learning. Its purpose is to cater to students with different learning preferences and avoid the “one-size-fits-all” approach. This model provides various suggestions for learning, and the current study incorporates these recommendations to effectively meet the diverse learning needs of students, as shown in Table 2. Thus, the learning objects are integrated with and added to these learning-style components to meet students’ learning demands.

Table 2.

Recommended Activities From FSLSM.

Learning style	Recommendations in learning
Active	Making a guess in possible questions and answering them
Reflective	Chances to write short summaries about the already learned material
Sensing	Facts, concrete material, and data and linear text
Intuitive	Facts and lesson objectives and linear text
Visual	Graphics, video, and images
Verbal	Text-based material
Sequential	Guidance
Global	Summaries

Source. Adapted from Graf (2007).

This study involved the development of a total of 60 learning objects for the LA intervention. Some of these learning objects incorporate elements that cater to more than one learning style. Table B1 provides a comprehensive overview of the LA intervention’s learning objects, categorized according to the ARCS model and FSLSM and aligned with various learning styles. Essentially, these learning objects were meticulously designed by incorporating the motivational components of the ARCS model along with additional learning style elements based on the FSLSM. Notably, an asterisk (*) denotes the inclusion of more learning objects tailored specifically to the learning styles of students identified as at risk in Step 3. To make these learning materials available to everyone, they were uploaded to the e-learning system, which allows all students to access them for free. Figure 2 provides an example, showing a learning resource that was developed by incorporating preferred learning styles and motivational elements before being uploaded to the online platform (Kew & Tasir, 2022a).

Figure 2.

Sample of learning object (LO).

After showing the learning style model components and motivational model components mentioned in the previous tables, which are integrated with the LA intervention in the form of learning objects, the table in Table C1 shows the instructional strategies in the e-learning environment embedded with the LA intervention. Three main topics were taught during the treatment period: Introduction to Microsoft Access Database, Application of Microsoft Access Database, and Introduction and Application of Open Source. Hence, three different learning objects of the LA intervention are developed based on these three main topics, namely learning objects 4 (LO4), learning objects 5 (LO5), and learning objects 6 (LO6), which are shown in Table C1, consisting of topic, subtopic, learning objectives, learning activities, learning objects, etc. Each LO has 20 small chunks of LOs: for example, LO4 has LO4_1, LO4_2, and so on. In this respect, a total of 60 LOs were developed for all three course topics.

In this intervened e-learning environment, students’ learning behaviors, such as the number of times they logged in, viewed content, and posted messages, were then measured, as well as whether this LA intervention was efficient in improving students’ motivation, academic results, engagement, and ability to retain what they learned in e-learning. Finally, using these results, a framework of students’ learning pathways was formulated.

Methodology

Research Design, Research Philosophy, and Research Approach

This study employed a one-group pre-test–post-test design to assess the impact of a Learning Analytics (LA) intervention in an e-learning environment. This design was selected due to its practicality in educational settings where random assignment is not feasible, especially since the sample in this study comprised the entire population. The intervention involved developing and implementing an LA system tailored to students’ learning styles and preferences, incorporating the Felder-Silverman Learning Style Model (FSLSM) and Keller’s ARCS model. Validated instruments were used to measure students’ motivation, engagement, and academic achievement before and after the intervention. The process included a pre-test (O1) to identify at-risk students, followed by the implementation of the LA intervention, and concluded with a post-test (O2) to evaluate the effects on students’ learning behaviors and performance. This design, though lacking a control group, is widely accepted in educational research for capturing within-subject changes (Jumaat, 2014). It allows researchers to observe the effect of the intervention by comparing students’ progress over time, offering insights into individual-level changes.

To strengthen the internal validity, the study combined quantitative and qualitative instruments, providing a more comprehensive understanding of the intervention’s impact. However, the design is subject to threats such as internal and external validity concerns. One internal threat is the history effect, where events occurring between the pre- and post-tests might influence results. To control for this, the intervention was conducted over 8 weeks, across two sessions before and after the mid-semester break, aligning with Jumaat’s (2014) similar approach. Another threat, maturation, was mitigated as students attended varying weekly lessons, reducing the likelihood of boredom or natural improvement. The focus on enhancing motivation, cognitive engagement, retention, and achievement aimed to ensure changes were due to the intervention rather than natural development. To address the selection-maturation interaction, all students received the same intervention, ensuring uniform exposure and reducing variability. In terms of external validity, the main concern is selection-treatment interaction, which limits generalizability if the sample is not representative. However, this threat was not present, as the study’s sample constituted the entire target population. Another external threat, the interaction of testing with treatment, refers to the potential influence of testing itself on outcomes. While it is acknowledged that pre-tests might impact results, this study aimed to assess improvement within the same group rather than generalize to others.

In summary, although the one-group pre-test–post-test design has limitations, it remains a robust approach in contexts where control groups are not viable. By carefully addressing validity threats and focusing on tailored interventions based on learning styles, this study aimed to evaluate the effectiveness of LA in improving student outcomes in an e-learning environment. Lastly, the total duration of the intervention is one full semester, which consists of 17 weeks. The collected data were used to create a framework that outlines the learning paths of learners with various learning styles in e-learning, enhanced by the LA intervention. This design allowed the researchers to measure changes in variables over time and understand individual performance alterations. Figure 3 outlines the research procedure, encompassing six phases.

Figure 3.

Research procedure.

This study adopts a pragmatic research philosophy, focusing on practical solutions to real-world problems, specifically enhancing student learning through Learning Analytics (LA) interventions. Pragmatism supports the use of both quantitative and qualitative methods to understand how learning styles influence engagement and performance in e-learning. A deductive research approach was employed, beginning with established theories on learning styles, engagement, and LA, followed by empirical testing. The study aimed to determine whether a tailored LA intervention could improve motivation, engagement, and academic performance, thereby validating or refining a theory-driven framework based on data and structured analysis.

Population and Sample

The study centers around the complete cohort of second-year undergraduate students enrolled in a computer-based course in one Malaysian university, serving as the population in this research. All 50 students from an educational background were selected as participants. Their ages ranged between 19 and 22 years old. Out of 50 participants, 28 were female, and 22 were male. Notably, all participants had prior experience with e-learning since their first year and demonstrated active engagement in e-learning activities. A power analysis is not necessary because this study involves the entire population.

Instruments and Data Analysis

Both Index of Learning Styles (ILS) and automatic detection of learning style were used to analyze the learning style of students. Firstly, the ILS is a questionnaire with 44 questions designed to identify people’s preferences in learning styles, based on Felder and Silverman’s model. It covers four areas: active versus reflective, visual versus verbal, sensing versus intuitive, and sequential versus global. Students completed this questionnaire, and the results were compared to an automated method for detecting learning styles that looked at students’ behaviors. To validate the instrument, in the pilot study stage, another new group of 42 of the undergraduate students was chosen to voluntarily answer this questionnaire, and the resulting Cronbach Alpha value for the ILS is .78, which is valid and reliable. Besides that, automatic detection of learning style was also used to analyze the learning style of students. The literature-based approach proposed by Graf et al. (2008) uses students’ behaviors to obtain hints about their learning style preferences. From the number of matching hints, a simple rule-based method is utilized to determine learning styles. This approach is actually parallel to the method used in the ILS questionnaire. By totaling all hints and dividing them by the number of patterns that comprise available information, a measure for the respective learning style is calculated, and is then normalized to identify learning styles for each dimension of the Felder-Silverman model. Based on previous research, the precision for all the dimensions of the Felder-Silverman model of the proposed method compared with the ILS questionnaire ranges from 73.33% to 79.33%, proving a promising use in determining the learning styles of students in an online environment. This research used a literature-based method to evaluate learning styles automatically with reference to the previous study conducted by Dung and Florea (2012). Timeexpected_stay and Timespent are examined in this research. Timeexpected_stay is described as the time expected to be spent on each LO; whereas, Timespent is explained as the time that a student actually spends on each LO. These values were measured for each learning style labeled for the LOs. In addition, the sum of Timespent is calculated for the learning style elements of the learner after a period P. Eight respective ratios are found:

$$\mathrm{R}{\mathrm{T}}_{1\mathrm{s}-\mathrm{element}}=\mathrm{\Sigma }\mathrm{Time}\mathrm{\_}\mathrm{spent}/\mathrm{\Sigma }\mathrm{Time}\mathrm{\_}\mathrm{expected}\mathrm{\_}\mathrm{stay}$$

The same method was used to find out the ratio RVLS_element, which is defined as the number of LOs visited. Then, in relation to each learning style element, the number of LOs visited and the total number of LOs were calculated as follows:

$$\mathrm{R}{\mathrm{V}}_{1\mathrm{s}-\mathrm{element}}=\mathrm{\Sigma }\mathrm{LOs}\mathrm{\_}\mathrm{visited}/\mathrm{\Sigma }\mathrm{LOs}$$

Finally, the average ratio was calculated: Ravg = (RT + RV)/2

Learning styles are measured and expected based on the following simple rule: Ravg values of 0 to 0.3, 0.3 to 0.7, and 0.7 to 1 indicate that learning style preferences are weak, moderate, and strong, respectively. The mutual outcome for two elements of learning style of the same dimension (which are both strong) was rejected.

Secondly, J. M. Keller’s (1987) Instructional Material Motivational Survey (IMMS) was used to measure students’ motivation levels. The 36-item survey includes four subscales: Attention (ATT), Relevance (RELE), Confidence (CONF), and Satisfaction (SAT). Responses were recorded on a 5-point Likert scale, where scores below 3.00 indicate low motivation, 3.00 to 3.49 medium, 3.50 to 3.99 upper medium, and 4.00 to 5.00 high motivation. Students completed the IMMS before and after the intervention. To validate the instrument, a pilot study with 36 undergraduate students was conducted. Exploratory Factor Analysis (EFA) showed strong Kaiser-Meyer-Olkin (KMO) scores: ATT (0.719), CONF (0.750), RELE (0.633), and SAT (0.747), all above the .6 threshold (Kaiser, 1974). Bartlett’s Test of Sphericity was significant (p = .000) for all subscales. The overall Cronbach’s alpha was .96, indicating high reliability and validity.

Thirdly, to assess cognitive engagement, online discussion forum scripts were collected and coded using Van der Meijden’s (2005) scheme. Cognitive engagement was determined by comparing the proportion of high- and low-level contributions. The coding process’s reliability was established through inter-rater reliability between the researcher and an expert. The inter-rater agreement was 87.93%, with a Cohen’s Kappa value of .951, indicating strong consistency. The discussion topics were validated by an expert with over 5 years of experience in designing online discussion forums.

Fourthly, cognitive retention was measured through students’ test scores. Two pre- and post-tests were conducted, each with a 3-week interval. Each test set consisted of three short essay questions (10 marks total). The second set used different questions from the first. Content validity was ensured by two experts with over 5 years of teaching experience. Reliability was confirmed through test-retest involving 10 students; the Pearson correlation was .761, which is considered reliable (Leech et al., 2011).

Fifthly, academic performance was evaluated using pre- and post-tests consisting of multiple-choice and essay questions. The tests were based on the Information System Management in Education course and validated using past exam questions. Each test included Section A (15 multiple-choice questions) and Section B (3 short essays), totaling 30 marks. Different sets of questions were created for each test session, aligned with specific topics. Scores were converted to percentages and assessed using the university’s grading system. Two experts validated the content, and reliability was confirmed via test-retest with 10 students. The Pearson correlation was .918, confirming strong reliability (Leech et al., 2011).

Students’ server log data from Moodle were also analyzed to understand behavioral patterns related to different learning styles in the LA-embedded e-learning environment. Data included the number of log-ins, activity views, and learning interactions (e.g., discussion posts). These behavioral metrics were used to develop a learning pathway framework integrating the LA intervention and student learning styles.

Lastly, to construct the learning pathways, data mining was conducted using WEKA 3-7-12 software. A decision tree algorithm with 10-fold cross-validation was applied, ensuring predictive accuracy. This analysis helped establish relationships between the LA intervention and variables such as learning styles, motivation, academic achievement, cognitive engagement, cognitive retention, and learner behavior. The resulting learning pathway framework identifies how students with different learning styles navigate e-learning environments and how LA interventions can be personalized to enhance their learning outcomes. The framework highlights the importance of aligning LA with learning styles to improve motivation, engagement, retention, and academic performance in e-learning.

Ethical Considerations

In this research, ethical and privacy concerns were prioritized through strict adherence to data protection principles. Informed consent was obtained by providing participants with a clear consent form detailing the research purpose, data collection methods, and their rights, ensuring they understood the procedures and could withdraw at any time without penalty. All personal identifiers, including names and contact information, were anonymized by using unique identification numbers (e.g., ID 1, 2, etc.), and data was aggregated to ensure privacy. Data was securely stored in password-protected files, with access restricted to authorized researchers. Ethical considerations were further upheld by obtaining permission from the course lecturer. Throughout the research, participants’ confidentiality was maintained, and their involvement had no impact on their academic outcomes. These measures ensured that the research was conducted with respect, fairness, and transparency.

Results

After students had been given the LA intervention, the students’ overall learning performance was examined (RQ1) to provide an overview of how the LA intervention assisted students in performing better in e-learning. Students’ learning behaviors based on learning style in terms of total number of log-ins, views, and posts were also examined (RQ2) to understand how students behaved in e-learning. With these results and data, a framework of learning pathways of students with various learning styles in an e-learning environment embedded with the LA intervention was formulated (RQ3).

Analysis of Learning Performance of Students Such as Motivation, Academic Achievement, Cognitive Retention, and Cognitive Engagement

From Table 3, the study showed that students’ overall motivation level increased slightly after receiving the LA intervention, from an average of 3.59 to 4.05. In other words, the students’ average motivation level went up. After the Shapiro-Wilk normality test, as the data were not normally distributed, a Wilcoxon test was used. The Wilcoxon test was conducted to compare the mean difference in students’ motivation scores before and after the LA intervention (Z = −5.954, p = .000). Hence, there is a significant difference in the average motivation levels before and after the LA intervention. The effect size for this study, based on Pallant’s (2007) method of dividing the test statistic by the square root of the number of observations, is .59. The Wilcoxon Signed Rank Test revealed a statistically significant increase in motivation after the LA intervention was provided (z = −5.954, p < .001), with a large effect size (r = .59) according to Cohen (1988). It shows that the LA intervention can help boost students’ motivation levels in e-learning.

Table 3.

Students’ Motivation Levels, Academic Achievement, Cognitive Retention, and Cognitive Engagement Levels Before and After the LA Intervention.

Variables	Before intervention		After intervention		Statistical test
	Mean	Std. deviation	Mean	Std. deviation
1. Motivation level (n = 50)	3.59	0.18	4.05	0.31	Wilcoxon Signed-Rank Test of motivation level Asymp. Sig. (2 tailed): .000
2. Academic performance (n = 50)	34.28	8.60	88.56	4.30	Paired sample t-test pre and post-test scores Sig. (2-tailed): .000
3. Cognitive retention (n = 50)	82.60	8.88	90.20	4.62	Wilcoxon Signed-Rank Test of cog. Retention Asymp. Sig. (2-tailed): .000
	Number of students	Percentage	Number of students	Percentage
4. Cognitive engagement:
i. High level	16	32.00	48	96.00
ii. High-low level	17	34.00	0	0.00
iii. Low level	17	34.00	2	4.00
Total	50	100.00	50	100.00

Their academic achievement in the subject they were taught was also analyzed. The mean pre-test score stood at 34.28%, whereas the mean post-test score rose to 88.56%. The difference in the mean scores demonstrates that students’ academic performance is enhanced after providing intervention. After the normality test, the paired-samples t-test for pre- and post-performance tests was conducted. The finding showed the significance value is .000. The effect size was calculated using Cohen’s d formula, which gives a value of 5.669. The power statistical test was run using standalone software called G*Power, and the power value for this analysis is 1 (p > .05). It shows that the LA Intervention can enhance students’ academic achievement in e-learning.

Besides that, the results indicate that the average cognitive retention pre-test score was 82.60%. After the implementation of the LA intervention, the average cognitive retention post-test score rose to 90.20%. This discrepancy suggests that students retained the knowledge imparted to them. The normality test was then conducted. As the data was not normally distributed, the Wilcoxon Signed Rank Test was conducted. As the Wilcoxon Signed Rank Test revealed a statistically significant increase in cognitive retention level in cognitive retention post-test 2, z = −5.390, p < .001, with a large effect size (r = .539), it shows that the LA intervention helped students to enhance their cognitive retention.

Moreover, the count of students exhibiting high-level cognitive engagement surged from 16 to 48 following the LA intervention. In contrast, those demonstrating low-level cognitive engagement dwindled from 17 to 2. This suggests the students put in more mental effort and actively participated in the discussion forum. Additionally, Table D1 provides a list of students with details on their motivation level, academic achievement, cognitive retention, and cognitive engagement based on their learning style. This reveals whether their learning performance improved, stayed the same, or declined after receiving the LA intervention.

Analysis of Learning Behaviors of Students in Terms of Log-ins, Views, and Posts

To understand how students with different learning styles use e-learning integrated with the LA intervention, the descriptive statistics of learner behaviors were used to measure the overall behavior of students in terms of log-ins to the system (number of log-ins), viewing activities by students (number of views), and interaction for learning (number of posts).

Table 4 and Figure 4 show that students with different learning styles behaved and engaged with the e-learning differently. From Table 4, in terms of log-ins to the system, students with a verbal learning style contributed most to this activity with the highest percentage (17.65%), as they logged into the e-learning environment the most frequently. However, regarding students’ viewing activities and interactions for learning, students with an active learning style contributed the most, with 20.49% and 18.38%, respectively. In other words, they viewed the e-learning materials, such as learning objects, more and interacted more enthusiastically with their peers in the discussion forum, as they felt motivated to view and post due to the effect of the LA intervention with personalized learning objects.

Table 4.

Average Number of Student Learning Behaviors in E-Learning After the LA Intervention Based on Types of Learning Style.

Learning style	n	Log in the system		Viewing activities by students		Interaction for learning
		Number	Percentage	Number	Percentage	Number	Percentage
Active	2	36	16.29	532	20.49	34	18.38
Reflective	5	34	15.38	393	15.13	27	14.59
Intuitive	6	38	17.19	433	16.67	28	15.14
Visual	31	23	10.41	300	11.55	26	14.05
Verbal	2	39	17.65	392	15.09	29	15.68
Sequential	1	27	12.22	205	7.89	13	7.03
Global	3	24	10.86	342	13.17	28	15.14
Total	50	221	100.00	2597	100.00	185	100.00

Figure 4.

Average number of student learning behaviors in e-learning after the LA intervention based on types of learning style.

Framework of Students’ Learning Pathway in an E-Learning Environment Embedded With Learning Analytics Intervention Based on Students’ Learning Styles

The decision-tree method using WEKA 3-7-12 data mining software was employed to construct a framework of students’ learning pathways in an e-learning environment that incorporated LA intervention according to students’ learning styles. The algorithm underwent a 10-fold cross-validation process. This means that it was given an opportunity to make a prediction for each instance of the dataset, and the presented result is a summary of those predictions. The framework was developed based on students who had demonstrated enhanced motivation, academic achievement, cognitive engagement, and cognitive retention in the e-learning environment with the LA intervention tailored to their learning styles (see Table D1). Therefore, this framework depicts the learning pathway of those students who experienced improvements in motivation, academic performance, cognitive engagement, and cognitive retention through the LA intervention integrated into their e-learning experience. Table 5 shows the data input used in the decision tree and its description.

Table 5.

Name of Data Input and Its Description.

No.	Name of data input	Description
1	PreferredLoBeingVisited Numeric	Total number of preferred LOs visited
2	ViewOnPreferredLO Numeric	Total number of views of preferred LOs
3	TotalPost Numeric	Total number of posts
4	ViewOnPost Numeric	Total number of views of posts
5	AcademicAchievement	The grade of academic achievement

Below shows the attributes used in the Decision Tree for building learning paths of students based on different learning styles.

• @Relation FrameworkLearningPathway

• @attribute PreferredLoBeingVisted Numeric

• @attribute ViewOnPreferredLO Numeric

• @attribute TotalPost Numeric

• @attribute ViewOnPost Numeric

• @attribute AcademicAchievement {A+,A,A−}

• @data

Data are mined using a decision tree according to each category of learning style, such as active, visual, global, reflective, and intuitive students, which are the types of learning style possessed by students who showed they had enhanced all motivation, academic achievement, cognitive engagement, and cognitive retention in e-learning. Nonetheless, students with active, global, reflective, and intuitive learning styles were not analyzed independently because these learning style categories had limited data sets that were insufficient to generate a learning pathway. As shown in Table D1, the active learning-style had only one student, while the global learning style had two students, the intuitive learning style had three students, and the reflective learning style had two students. In this case, therefore, only students with a visual learning style were involved in generating the learning pathways.

Thus, the following sections show two learning pathways, which were generated based on two types of visual students, to show how the visual students performed in an e-learning environment embedded with the LA intervention based on students’ learning styles:

1. The first group (n = 15) comprised those visual students who showed enhanced scores in all aspects: motivation, academic achievement, cognitive engagement, and cognitive retention, and

2. The second group (n = 16) showed enhanced scores in some but not all aspects of motivation, academic achievement, cognitive engagement, or cognitive retention.

Learning Pathway of Visual Students Who Had Enhanced All Motivation, Academic Achievement, Cognitive Engagement, and Cognitive Retention

Table 6 shows the log data of visual students who showed enhanced scores in all of their motivation, academic achievement, cognitive engagement, and cognitive retention, including the total number of posts, the total number of preferred LOs being visited, the total number of views of posts, and the total number of views of preferred LOs. A framework of students’ learning pathways was generated based on these students’ log data as collected from the e-learning environment.

Table 6.

Log Data of Students Who Had Enhanced All of Their Motivation, Academic Achievement, Cognitive Engagement, and Cognitive Retention.

ID	Learning style	Total number of view of posts	Total number of preferred LOs being visited	Total number of posts	Total number of views on preferred LOs	Academic achievement
2	visual	76	5	16	66	A+
6	visual	78	9	34	93	A+
15	visual	31	9	9	78	A+
18	visual	144	12	31	50	A
25	visual	136	8	32	100	A+
30	visual	171	15	62	177	A+
31	visual	117	10	44	53	A+
37	visual	79	13	20	98	A
40	visual	110	14	38	101	A+
41	visual	148	13	29	98	A+
45	visual	140	14	39	105	A−
46	visual	58	3	16	28	A
47	visual	110	6	26	97	A
48	visual	50	15	19	129	A+
50	visual	166	8	46	207	A+

Note. n = 15.

Firstly, Figure 5 shows the result of the learning pathway of these visual students. It was found that six attributes contributed to the formation of the learning pathway for these students, namely, ViewOnPreferredLO, ViewOnPost, PreferredLoBeingVisited, A+, A, and A−. As depicted in Figure 4, several patterns emerged for five types of learning pathways, which are academic achievement of A+, A, and A−, as shown in Table 7. Two types of learning pathways for A+ and A were found based on the students’ academic achievement.

Figure 5.

Learning pathway of visual students who had enhanced all motivation, academic achievement, cognitive engagement and cognitive retention.

Table 7.

Learning Pathway of Visual Students Who Had Enhanced All Motivation, Academic Achievement, Cognitive Engagement, and Cognitive Retention.

Academic achievement	Pathway
1. A+ Type I (8 students)	ViewOnPreferredLO → ViewOnPost → PreferredLoBeingVisited
2. A+ Type II (2 students)	ViewOnPreferredLO → ViewOnPost → ViewOnPost
3. A Type I (3 students)	ViewOnPreferredLO
4. A Type II (1 student)	ViewOnPreferredLO → ViewOnPost → PreferredLoBeingVisited
5. A− Type I (1student)	ViewOnPreferredLO → ViewOnPost → ViewOnPost

The outcome of the decision tree indicates that the correctly classified instances stand at 73.33%, with a Kappa statistic of .4595 and a mean absolute error of 0.1778. The precision of the predicted model to the actual model is with an ROC area of 0.729. According to Table 8, the reason A− was not predicted correctly was because it has only one student; therefore, it was unable to do a comparison.

Table 8.

Confusion Matrix.

Items	Results
Confusion matrix	a	b	c	classified as
	8	0	2	a = A+
	1	3	0	b = A
	1	0	0	c = A−

Learning Pathway of Visual Students Who Had Enhanced Some But Not All of Motivation, Academic Achievement, Cognitive Engagement or Cognitive Retention

Table 9 shows the log data of visual students who had enhanced some but not all of their motivation, academic achievement, cognitive engagement, or cognitive retention. Another framework of students’ learning pathway was generated based on these students’ log data.

Table 9.

Log Data of Visual Students Who Had Enhanced Some But Not All of Motivation, Academic Achievement, Cognitive Engagement, or Cognitive Retention.

ID	Learning style	Total number of posts	Total number of preferred LO being visited	Total number of views on post	Total number of views on preferred LO	Academic achievement
1	visual	10	7	17	37	A+
8	visual	8	6	16	52	A+
9	visual	15	5	53	40	A+
14	visual	8	4	17	29	A+
16	visual	16	6	59	53	A+
21	visual	11	5	43	23	A+
23	visual	9	8	79	124	A+
26	visual	57	10	130	102	A+
28	visual	16	10	49	74	A+
38	visual	16	16	41	133	A+
42	visual	28	6	124	62	A+
44	visual	34	8	111	93	A+
51	visual	22	3	47	18	A+
11	visual	19	3	37	17	A
20	visual	26	11	56	29	A
32	visual	46	6	139	36	A

Note. n = 16.

Corresponding with this, Figure 6 shows the result of the learning pathway for those students with a visual learning style. Five attributes were found to contribute to the formation of the learning pathway for these students, namely, PreferredLoBeingVisited, TotalPost, ViewOnPreferredLO, A+, and A.

Figure 6.

Learning pathway of visual students who had enhanced either motivation, academic achievement, cognitive engagement, or cognitive retention.

As depicted in Figure 6, several patterns emerged for four types of learning pathways, which are academic achievement of A+ and A, as shown in Table 10.

Table 10.

Learning Pathway of Visual Students Who Had Enhanced Some But Not All of Motivation, Academic Achievement, Cognitive Engagement, or Cognitive Retention.

Academic achievement	Pathway
1. A+ Type I (12students)	PreferredLoBeingVisited → ViewOnPreferredLO
2. A+ Type II (1 student)	PreferredLoBeingVisited → TotalPost
3. A Type I (2 students)	PreferredLoBeingVisited → TotalPost
4. A Type II (1 student)	PreferredLoBeingVisited → ViewOnPreferredLO

Moreover, from Table 11, it shows the correctly classified instances are 87.5%, the Kappa statistic is .5897, and the mean absolute error is 0.125. The precision of the predicted model to the actual model is with a ROC Area of 0.795.

Table 11.

Confusion Matrix.

Items	Results
Confusion matrix	a	b	classified as
	12	1	a = A+
	1	2	b = A

Lastly, based on the results of the two learning pathways for visual students, a framework of learning pathways was developed. This framework shows how visual students, whose motivation, academic achievement, cognitive engagement, and cognitive retention were enhanced, behaved in e-learning environments that incorporated the LA intervention. The framework is shown in Figure 7. For learning pathway 1, 15 students had fully enhanced their motivation, academic achievement, cognitive engagement, and cognitive retention. They had identified their preferred learning objects and posted in the discussion form. On the other hand, learning pathway 2 is for students who had partially enhanced their motivation, academic achievement, cognitive engagement, and cognitive retention in e-learning embedded with the LA intervention. From this learning pathway, most of the preferred learning objects were visited by students, and they posted more in the discussion forum.

Figure 7.

Framework of learning pathway of students with visual learning style in e-learning environment embedded with the LA intervention.

Figure 7 indicates that there are two learning pathways of visual students, which show (1) full (which means all motivation, academic achievement, cognitive engagement, and cognitive retention) and (2) partial (which means either motivation, academic achievement, cognitive engagement, or cognitive retention) enhancement in e-learning. In order to fully enhance all of their learning performance by using the LA intervention, it was found that 60% of students had viewed their preferred LOs and the posts in the discussion forum many times. They had also gone through the LOs provided for them, which were designed to meet their learning preferences based on their learning style. Twenty percent of them viewed their preferred LOs, while the other 20% viewed their preferred LOs and the posts in the discussion forum. On the other hand, 81% of students who had gone through the LO provided for them and viewed these LOs had enhanced some but not all of their motivation, academic achievement, cognitive engagement or cognitive retention in e-learning, whereas, 19% of them had gone through the LO provided and posted messages in the discussion forum to achieve the same performance as the aforementioned 81% of students.

Discussion

The study findings show the LA intervention significantly boosted students’ learning. Students’ average test scores jumped from 34.28 to 88.56, indicating the LA intervention can help address learning challenges in e-learning. This aligns with results from the Course Signals and OAAI projects, suggesting LA-based interventions can effectively improve students’ course performance. Overall, the LA intervention enhanced students’ motivation, academic achievement, cognitive engagement, and retention in e-learning. This finding is primarily related to the benefits of early identification of students who are at risk of struggling, and then designing meaningful learning activities for the LA intervention. This intervention is integrated with a model that considers the students’ learning styles and motivational factors, with the goal of helping them enhance their learning performance. Secondly, the study revealed positive changes in students’ motivation over the course. The majority of students were motivated and satisfied with the personalized learning materials provided through the LA intervention. This suggests that well-designed learning objects can effectively maintain student motivation. The findings show the students’ overall motivation level increased from an average of 3.59 to 4.05 after the LA intervention, which encouraged them to continue learning in an environment that catered to their preferred learning styles.

Thirdly, the results indicate that the number of students experiencing increased cognitive engagement rose from 16 to 48 following the implementation of the LA intervention. This directly suggests that during forum discussions, 48 students achieved elevated levels of knowledge construction, entailing heightened cognitive effort in elaborating on facts and arguments. Prior research has established a close link between cognitive engagement and student motivation. As discussed in the preceding section, the findings demonstrate an increase in student motivation levels, which subsequently contributed to enhanced cognitive engagement. Lastly, the results demonstrated an improvement in overall cognitive retention among students’ post-intervention, indicating that the intervention facilitated more effective information retention. Sustaining students’ motivation, engagement, and retention is imperative in the e-learning process to optimize learning outcomes for all students.

Besides that, based on the data collected from students’ activities in the e-learning environment, such as their log-ins, posts, and views of learning materials, the results show that students with different learning styles had varying engagement levels. Active learners had the most views and posts, while verbal learners logged in the most. This suggests that the LA intervention, which provided learning objects tailored to their preferences, positively influenced their learning behaviors and increased their motivation to use the e-learning platform. Importantly, the results show that most students accessed the learning objects preferred by their individual learning styles. This indicates they tried to use the personalized learning objects suited to their preferences. The data logs confirm that the students’ learning behaviors and responses were influenced by the e-learning environment with the LA intervention. The e-learning intervention was able to cater to students with different learning styles by providing their preferred learning objects.

Furthermore, it is also important to understand what students’ learning pathways look like in e-learning, so that the pathway enables each student to achieve the best learning outcome possible. From the framework developed for students who (1) enhanced all four aspects – motivation, academic achievement, cognitive engagement, and cognitive retention – it was found that the total number of views of preferred learning objects for this learning style, total number of views of posts, and total number of preferred learning objects being viewed are the central elements of such a framework. In other words, the number of views of preferred learning objects matched with the visual learning style was significant in determining the pathway for visual-learning-style students who enhanced their motivation, academic achievement, cognitive engagement, and cognitive retention in the e-learning environment embedded with the LA intervention. It can be clearly seen that learning objects matched with the visual learning style are helpful to enhance the learning performance for these students in an e-learning context.

This appears to be consistent with the findings reported by Perera and Richardson (2010) in that the more students view and read a learning object, the more they tend to learn. This also confirms that the learning objects preferred by students with a visual learning style, such as pictures, graphics, and videos, were extremely useful for these students and attracted their attention. It can be concluded that the total number of preferred learning objects being viewed plays another significant role, and visual students are strongly encouraged to view the learning objects preferred by their learning style, as they can easily remember what they have seen to achieve better examination results. They also viewed other students’ posts to verify their understanding and construct their knowledge in the discussion forum. By doing this, they may achieve better results. Thus, it is important for instructors to collect viewing data each time a student visits the e-learning environment to understand which learning objects were viewed and help them learn. Balakrishnan and Coetzee (2013) computed the number of objects viewed per week for every student and argued that this is an important step to improve students’ learning performance.

On the other hand, the central elements of the framework developed for students who (2) enhanced some but not all of their motivation, academic achievement, cognitive engagement, or cognitive retention are the total number of preferred learning objects being viewed, the total number of views of preferred learning objects by this learning style, and the total number of posts. In other words, these elements play important roles in the learning pathway of students to enhance either motivation, academic achievement, cognitive engagement, or cognitive retention in an e-learning environment embedded with the LA intervention. Apart from the importance and impact of the total number of preferred learning objects being viewed and the total number of views of learning objects preferred by each learning style, which were explained previously, another learning behavior change that should be considered in this second learning pathway is the creation of posts, which might involve asking questions, responding to questions, or commenting on topics. This view is in line with the findings reported by Romero et al. (2013), who found that participation in online discussion forums, such as the number of posts, is related to a student’s final grade. This might be because they were curious about which topics their fellow students emphasized and what they learned from the learning objects, and they were also curious about new posts or replies by their friends. This finding also suggests a direct interactive relationship among students and concurs with other findings, which indicate that LA interventions can support changes in students’ involvement in discussions (Wise et al., 2014).

Furthermore, the developed framework centered around visual learners was guided by the Felder-Silverman model. Abrar et al. (2025) explore the effectiveness of AI-powered smart learning paths and dynamic assessments in enhancing learning efficiency by personalizing education based on individual student performance, preferences, and learning behavior. The proposed AI-driven model adapts the flow, content, and modality of instruction through continuous data collection, machine learning analysis, and real-time feedback. It dynamically adjusts learning materials to match each student’s level, offering adaptive content, customized formats (e.g., visual, textual, action-based), and tailored assessments that evolve based on progress. Empirical results demonstrate that the intervention led to a 25% improvement in performance, 25% faster task completion, and a 15% increase in student engagement compared to traditional methods, highlighting the model’s potential to significantly enhance learning outcomes and motivation. Although Abrar et al.’s study incorporates customized formats aligned with learning preferences, it uses AI and not an LA intervention to support students’ learning styles. Therefore, the development of learning pathways, especially for visual learners, is important, and this research serves as a valuable reference for future studies aiming to enhance personalized learning through adaptive technologies. This personalized approach keeps learners engaged and provides targeted support. While this study focused on visual learners, the methodology employed, including the use of learning analytics, motivational scaffolding with Keller’s ARCS model, and data mining techniques, is adaptable to other learning styles, given appropriate learner profiling. Despite the promising results of the Learning Analytics intervention framework, its implementation in real-world educational settings presents several challenges. For example, integration with existing Learning Management Systems can be technically complex, as many platforms may not support adaptive features, and instructors may face difficulties interpreting analytics data due to limited data literacy training and increased workload demands.

Conclusions, Limitations, and Recommendations for Future Studies

This study has shown that the implementation of the LA intervention in e-learning can help enhance the quality of education by meeting the needs of students with different learning styles. Also, it shows that LA intervention can help to solve learning issues in e-learning. In particular, the overall learning performance of students was improved, and students behaved and engaged in different ways; thus, RQ1 and RQ2 are answered. It also shows that the process of identifying at-risk students and then selecting and providing personalized learning objects through the LA intervention suitable to these types of learners is needed. Therefore, this study adds to the current understanding of how LA can be utilized to enhance students’ learning experiences and refine their learning environments, particularly in accommodating various learning styles. It addresses a research gap by focusing on the development of interventions to support students, especially those identified as at-risk, within the research field of LA. Moreover, a framework of the learning pathway of visual students in e-learning was formulated based on this information, which has answered RQ3. Key values, such as the total number of preferred learning objects being viewed, the total number of views of preferred learning objects by each learning style, and so on, are important for the efficiency of the learning process to achieve better academic achievement among visual students. The learning pathway highlights that the inclusion of visual learning objects in e-learning supports visual learners, who remember best what they see. It may provide a good reference point for visual students to follow this learning pathway to achieve better results by viewing the preferred learning objects and being engaged in discussion forums. Thus, it is important to implement LA interventions in the learning process, and instructors should try to use such interventions, which could increase the learning efficiency for this group, and take the findings of this research into consideration when designing, developing, and delivering learning objects to enrich students’ learning process in e-learning.

This research has some limitations. Although this research has developed a new LA intervention after identifying at-risk students in a detailed way, followed by the research on examining the effectiveness of this new LA intervention on students’ learning performance and learning behavior in the e-learning environment integrated with LA intervention, and lastly developing a framework of the learning pathway that was not formulated in other studies, the population in this research was only 50 undergraduate students and they were from the same course. Hence, further research can explore the effectiveness of an LA intervention in other disciplines or subject areas, and greater numbers of students should be used. Nonetheless, these 50 undergraduate students were investigated in depth in e-learning. Besides that, more different LA interventions should be developed to help students with different learning demands. This is because students have diverse backgrounds and different learning problems in their learning process. Even though this research has developed the LA intervention to cater to students with different learning styles, in order to solve the one-size-fits-all problem in online learning, future research should design more different interventions for different students after identifying the problems encountered by them in online learning. In addition, potential challenges in real-world settings, such as integration with existing e-learning systems and the need for educator training and support are acknowledged. These considerations provide a realistic perspective on applying the proposed framework; however, such challenges can be addressed with adequate institutional support.

Furthermore, LA is an emergent research field that has brought benefits to the educational field (Kew & Tasir, 2022b), for instance, enabling the evidenced-based decision to identify at-risk students and provide suitable intervention for them to meet their learning needs and improve the quality of teaching. It also helps to provide the hidden information about students, such as how they learn, behave, engage, and perform in e-learning. However, this research is only able to generate a framework of students’ learning pathways in an e-learning environment embedded with the LA intervention for the visual learning style. LA serves as a basis for interventions to cater to students with different learning styles. There are very limited previous studies or assumptions on learning styles, and in this research, when generating the framework for the students’ learning pathways, only visual-learning-style students provided sufficient log file data for building the framework of learning pathways in e-learning. This is because the data sets for the other learning styles were inadequate; nonetheless, this research serves as a valuable starting point and reference for future studies, as no similar research has been conducted to date. Future research should involve additional learning style groups with larger and more balanced data sets. It is also recommended that future studies explore the effectiveness of Learning Analytics interventions in other disciplines or subject areas, particularly those with more diverse demographic profiles. Moreover, further investigations could incorporate emotional data in e-learning environments to gain deeper insights into students’ behavior and backgrounds, and include control groups for comparative analysis. Lastly, future research should consider employing a broader range of data mining techniques, such as clustering, association rules, and classification, to uncover different patterns in students’ learning behaviors within e-learning contexts.