Time Series Forecasting on OULAD

Forecasting student interactions with the Open University's Virtual Learning Environment using four modeling approaches: SARIMA, ARIMAX, Prophet, and CNN. The CNN achieved the best MAE and MAPE across all evaluation windows.

Team

• Parham Khosh Solat
• Farshad Farahtaj
• Zahra Jafarinejad

Approach

The OULAD dataset tracks weekly student clicks on VLE materials across multiple courses and presentations. The goal was to predict future interaction volume from historical patterns.

Each model handles time series differently. SARIMA and ARIMAX capture seasonality and external variables well; Prophet handles trend changes; the CNN learns temporal patterns directly from raw sequences. Lag features were engineered for ARIMAX and CNN — this measurably improved both.

All models were evaluated on MAE and MAPE to allow direct comparison.

Dataset

Open University Learning Analytics Dataset (OULAD) — CC BY 4.0

studentVle.csv exceeds 400MB and is excluded from this repo. Download from Kaggle and place in /data.

Project Structure

├── data/                  # Download from Kaggle
├── notebooks/
│   └── Harware_and_Software_Mod_B_Final_Project.ipynb
├── src/
│   └── Hardware_and_sofware_Mod_B_Final_Project_Sreamlit_version.py
├── models/
│   └── tuner0.json
├── docs/
│   └── Harware_and_Software_Mod_B_Final_Project.pdf
├── requirements.txt
└── README.md

Setup

git clone https://github.com/parhamkhoshsolat/time-series-OULAD.git
cd time-series-OULAD
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
jupyter notebook notebooks/Harware_and_Software_Mod_B_Final_Project.ipynb

To run the Streamlit app:

streamlit run src/Hardware_and_sofware_Mod_B_Final_Project_Sreamlit_version.py

Course

Hardware and Software for Big Data (Module B) — University of Naples Federico II Instructor: Prof. Flora Amato · July 2024

License

MIT