Deep Knowledge Tracing (DKT) – Student Mastery Prediction

Overview

This project uses Deep Knowledge Tracing (DKT), a deep learning-based approach to estimate a student's knowledge level from their past learning interactions.

The model uses a Long Short-Term Memory (LSTM) network to analyze the sequence of student responses and predict their probability of answering questions correctly across different skills.

The goal is to identify students' weak areas and support personalized learning through data-driven insights.

Key Features

• Predicts student mastery probabilities across multiple skills.
• Uses historical student interactions as sequential input.
• Learns patterns in student learning and performance using an LSTM network.
• Identifies weaker and stronger skills.
• Provides skill-level predictions with readable skill names.
• Supports integration with a FastAPI-based tutoring backend.

Model Architecture

The DKT model consists of:

1. Embedding Layer: Converts skill-response interactions into dense representations.
2. LSTM Layer: Learns sequential patterns from a student's learning history.
3. Fully Connected Layer: Generates prediction scores for all skills.
4. Sigmoid Activation: Converts prediction scores into probabilities.

Model Configuration

Parameter	Value
Model	Deep Knowledge Tracing
Architecture	LSTM
Number of Skills	123
Embedding Dimension	64
Hidden Dimension	128
Maximum Sequence Length	200
Optimizer	Adam
Learning Rate	0.001
Training Epochs	20
Loss Function	Binary Cross Entropy with Logits

Model