Research
Adaptive AI Tutor Research Lab
An applied machine learning research project that investigates adaptive tutoring through student correctness prediction, feature engineering, recommendation strategies, clustering, and reinforcement learning.
Explore projectOverview
The project started with a reproducible synthetic adaptive-tutor prototype and was extended to real educational data using the EdNet KT1 dataset. The real-data stage focuses on predicting whether a student will answer a question correctly from previous interaction history and question-level signals.
Problem
Educational systems generate large amounts of student interaction data, but turning that history into reliable personalized learning decisions remains challenging.
Approach
The research pipeline combines leakage-aware feature engineering, supervised learning baselines, real-data scale experiments, feature ablation, feature importance, error analysis, student clustering, recommendation strategies, and a simulated Q-learning tutor.
Results
EdNet 500k
ROC-AUC 0.7301
EdNet 200k
ROC-AUC 0.7222
EdNet 50k
ROC-AUC 0.7080
EdNet 5k
ROC-AUC 0.6250