Analysis and Personalization of Gamification Elements Using Machine Learning Based on User Characteristics

Abstract

Background

Gamification is widely used to enhance engagement, motivation, and performance across domains such as education, health, and business. However, one-size-fits-all designs often fail to address the multidimensional nature of user characteristics, resulting in inconsistent outcomes. Most personalization strategies focus on single traits (e.g., player type) and neglect personality, cultural, and persuasive factors. This study proposes a machine learning–based framework that integrates a comprehensive user model to predict and personalize eight categories of gamification elements.

Methods

Survey data from 260 participants were collected via questionnaires measuring demographics, personality traits (NERIS), cultural dimensions (Hofstede), persuasive strategies (Cialdini), and player types (HEXAD), along with preferences for eight gamification element categories. Preprocessing steps included discretization, SMOTE balancing, and ANOVA F-value feature selection. Six classifiers, Logistic Regression, Decision Tree, Random Forest, SVM, Naïve Bayes, and XGBoost, were trained and evaluated with 10-fold cross-validation. A gamified image-labeling application was implemented in two versions: a generic design and a personalized design based on model predictions. A between-subject experimental evaluation with 80 participants compared both versions on six engagement and satisfaction metrics.

Results

In the two-class prediction setting, Random Forest achieved the highest performance (accuracy = 0.852, F1-score = 0.855), while SVM slightly outperformed in the three-class setting (accuracy = 0.768, F1-score = 0.764). Gender consistently emerged as a top predictor, followed by personality traits, cultural dimentions, persuasive strategies, and player types. Experimental evaluation showed significant improvements for the personalized version across all six metrics, with gains of 41% (performance satisfaction), 40% (impact on satisfaction), 30% (appeal), 41% (enjoyment of personalization), 43% (encouragement for additional activities), and 31% (motivation).

Conclusion

Integrating multidimensional user characteristics with machine learning enables accurate prediction of gamification preferences and significantly improves user experience. This approach addresses the limitations of one-size-fits-all gamification, providing a scalable, data-driven framework for delivering personalized, impactful designs.