Abstract
This paper proposes an adaptive educational platform that leverages machine learning (ML) to address the lack of personalized feedback in preparing for Kazakhstan’s Unified National Test (UNT). A quantitative survey of 109 recent UNT takers and first-year Astana IT University students reveals a strong desire for real-time analytics, topic-specific practice, and interactive exam preparation, although moderate trust in ML remains a concern. Based on these findings, a decision-tree-based recommendation system is presented, classifying student competencies (weak versus strong) at the subtopic level by analyzing accuracy, response times, and question difficulty. The system then generates targeted study suggestions, reflecting local linguistic and infrastructural contexts. Preliminary analyses suggest that such an approach can enhance student engagement, support efficient revision schedules, and reduce skill gaps—provided challenges involving inconsistent internet access, linguistic diversity, and distrust in automated tools are carefully addressed. In this paper, it is further demonstrated how transparent recommendation logic and iterative pilot testing with local educators can ensure cultural responsiveness. By embedding real-time performance tracking and dynamic content personalization, this ML-driven platform aims to modernize UNT preparation while fostering greater educational equity in Kazakhstan.