CELI Planner

An explainable cognitive system for adaptive e-learning interventions using affective-cognitive reasoning. The project uses OULAD learner data to predict academic risk, infer cognitive learner states, select interventions using a Deep Q-Network simulation, and explain decisions with SHAP/LIME.

What this project demonstrates

• Feature engineering from educational interaction data
• Gradient Boosting risk prediction with AUC/accuracy evaluation
• Rule-augmented cognitive state reasoning
• Deep Q-Network based intervention selection
• Simulated intervention outcome estimation
• Explainability using SHAP and LIME

Dataset

This project uses the Open University Learning Analytics Dataset (OULAD), which contains student demographic, assessment, course, and virtual learning environment interaction data.

Dataset source: Kaggle - OULAD Open University Learning Analytics Dataset

Note: The raw dataset files are not included in this repository due to file size. Please download the dataset from Kaggle and place the required CSV files inside the data/raw/ directory.

How to run

1. Create a Python environment.
2. Install dependencies:

pip install -r requirements.txt

3. Add the OULAD CSV files to the data/ folder.
4. Open and run:

notebooks/CELI_Planner.ipynb

Main pipeline

1. Load OULAD CSV files.
2. Engineer learner features.
3. Train a Gradient Boosting risk prediction model.
4. Convert risk and behavior signals into cognitive learner states.
5. Train a DQN intervention policy in a simulated environment.
6. Generate explainable reports using SHAP and LIME.

Notes

The intervention effect is evaluated through a simulated environment, not through real-world deployment. This is suitable for an academic prototype but should be validated with real learners before practical use.