Gamification has come up as a powerful tool in increasing learning engagement and motivation in online education. As a proposal in this article, the designing and developing of AI-powered game modules in GameMaker for STEM education with adaptive learning capabilities in a gamified setup is ideal. The proposed model can regulate AI in a controlled manner to personalize learning experiences based on game gameplay with adaptive difficulty levels, immediate feedback, and analysis of performance data of learners. Based on developing a game using GameMaker, this research utilizes AI analytical capabilities for achieving learning outcomes in STEM education by aligning game targets with learning in STEM education. A quasi-experimental design was utilized to assign 60 purposively sampled STEM students into experimental and control groups in equal measures. The data gathering methods used in this study include pre and post-tests, in-game engagement metrics and usability surveys based on SUS and TAM. Salient findings show that the AI-enhanced module produced significantly greater learning gains (M=19.80) than the non-adaptive module (M =11.10). Improvements within both groups were significant (p < 0.001), while between-group comparison favored the experimental group (p=0.002). Engagement metrics further revealed that the AI-adaptive group exhibited longer time-on-task, higher level-completion rates, and more active use of hints-indicating sustained engagement and effective utilization of adaptive support.

This study aims to evaluate the developed predictive dropout risk model in the Alternative Learning System (ALS) by analyzing various demographic, socio-economic, academic, and behavioral factors. The early identification of students who are at risks in dropping out is crucial in order to provide necessary academic intervention programs. The researcher used Knowledge Discovery in Databases (KDD) as methodology in the evaluation of the predictive models. Using Gradient Boosting Decision Trees (GBDT) for predictive modeling. Key findings highlighted that with both classes achieving an F1-score of 93% which demonstrate a balanced performance between precision and recall for both positive and negative classes. In summary, the overall evaluation of the system is 3.59 which indicates that they system can be used for deployment and maybe further be improved.