Using Machine Learning for Identification: Development of the Cognitive Assessment Battery for Twice Exceptionality

Abstract

The identification of twice-exceptional (2e) students is a complex challenge, primarily due to cognitive masking. In this study, we developed the Cognitive Assessment Battery for Twice Exceptionality (2eCAB) and evaluated the classification performance of a machine learning algorithm, specifically Classification and Regression Trees (CART). Grounded in prior literature, the 2eCAB was designed to assess nonverbal ability, memory, rapid automatized naming, and pseudoword reading. The sample included 565 Turkish-speaking elementary students: typically developing (TD, n = 468), gifted (n = 44), 2e (n = 15), and students with specific learning disabilities (SLD, n = 38). The results indicated that the 2eCAB is a valid and reliable tool. Internal consistency of the battery was high (α = .95, ω = .95). Test–retest reliability for total scores was .92, while individual task scores ranged between .77 and .92. Significant relationships were found between 2eCAB scores, and results from hierarchical confirmatory factor analysis showed a good model fit. Scores from four external assessments measuring nonverbal ability, working memory, naming speed, and reading were significantly correlated with 2eCAB scores. The trained CART algorithm achieved an acceptable overall classification accuracy for identifying 2e, gifted, TD, and SLD students. Thus, artificial intelligence technologies show promise for the identification of students with special needs.

Plain Language Summary

How Machine Learning Can Help Identify Students With Both Gifts and Learning Challenges

Some students are both highly gifted and have learning difficulties, such as dyslexia. These students are referred to as “twice-exceptional” (2e). Because their strengths and challenges can mask each other, 2e students are often misunderstood or misidentified in schools. This study introduces a new assessment tool called the Cognitive Assessment Battery for Twice Exceptionality (2eCAB) to improve the accurate identification of these students. The 2eCAB measures four critical cognitive skills: nonverbal reasoning, memory, reading pseudowords, and rapid naming. A total of 565 elementary students in Türkiye participated, including typically developing students, gifted students, students with learning disabilities, and 2e students. To better understand and classify these students’ cognitive profiles, researchers used a machine learning method called Classification and Regression Trees (CART). Due to the small number of 2e students, the researchers carefully tested several different approaches to address this imbalance, repeating their analyses many times to ensure reliability. The results showed that the 2eCAB is reliable and valid for identifying different student groups. The best-performing machine learning approach accurately identified students into their correct groups with high accuracy. However, the study also highlighted that some common practices, like oversampling data before splitting into training and testing sets, can produce overly optimistic results and should be avoided. Overall, these findings indicate that artificial intelligence can effectively support teachers, psychologists, and other professionals in identifying students with complex learning profiles. Importantly, the study emphasizes that machine learning should complement, not replace, professional judgment. Combining AI results with teacher observations, academic performance, and comprehensive assessment data will lead to better-informed decisions about students’ education.