Education Platform Student Answer Prediction
Predicting student responses using Item Response Theory and machine learning models.
Check out the full report here.
This project involved predicting student answers on an educational platform by implementing a learning-based Item Response Theory (IRT) Model. The goal was to model student ability against question difficulty to formulate a probability distribution for correct responses.
Key Achievements:
- Item Response Theory Model: Developed 1-parameter and 3-parameter IRT models to predict student success based on question difficulty and student ability.
- Model Optimization: Experimented with Autoencoders, Matrix Factorization, Neural Networks, and Ensemble methods to enhance prediction accuracy.
- Data Processing & Accuracy: Achieved a final validation accuracy of 71.1% and test accuracy of 70.3% using the IRT model, improving accuracy by adding guessing and discrimination parameters.