- Authors: Shrijaa Venkatasubramanian Subashini, Rahiq Raees, Christine Chow & Jiro Amato
This is a data analysis project for DSCI 522 (Data Science workflows); a course in the Master of Data Science program at the University of British Columbia.
We built a linear regression model utilizing Ridge Regression to predict a student’s final grade (G3) in a Portuguese language course using demographic data, behavioural insights, and previous assessment scores. Our final model achieved strong performance on unseen test data, with an R² score of 0.857 and a Mean Absolute Error (MAE) of 0.776. On average, model predictions deviated from actual grades by less than one grade point on the 0-20 scale, indicating a high level of accuracy. Although the model demonstrated reliability in predicting grades within the passing range (scores exceeding 10), residual analysis revealed reduced precision for students with lower scores or those receiving a zero, frequently overestimating their performance. Given this limitation concerning failing students, we conclude that this model showcases adequate performance to serve as a decision-support tool for educators to project final grades midway through the term.
The dataset used in this project is the Student Performance dataset created by Paulo Cortez from the University of Minho, Portugal (Cortez and Silva 2008). It was sourced from the UCI Machine Learning Repository (Cortez 2014) and can be found here. The dataset contains information on 649 students from two Portuguese secondary schools, with data collected through school reports and questionnaires. Each row represents a student with 30 features including demographic information (age, sex, family size), educational background (parental education, past failures, study time), social factors (going out, romantic relationships, alcohol consumption), and school-related features (absences, extra support, desire for higher education). The dataset also includes grades from the first period (G1), second period (G2), and final grade (G3), with G3 serving as the target variable for prediction.
The final report can be found here.
If you are using Windows or Mac, make sure Docker Desktop is installed and running.
Clone this GitHub repository:
git clone https://github.com/jiroamato/student_grade_predictor.git- Navigate to the project root in the terminal and run the following command:
docker compose up- Open the JupyterLab URL displayed in the terminal
http://127.0.0.1:8888/labor simply typelocalhost:8888in your browser.
- Enter the following commands in the terminal inside the JupyterLab instance with project root as the working directory to reset the project to a clean state (i.e., remove all files generated by previous runs of the analysis):
make clean- To run the analysis in its entirety, enter the following command in the terminal in the project root:
make all- To stop the container, press
Ctrl+Cin the terminal and run:
docker compose down- To remove the image that was pulled locally, note the image name and tag from
docker-compose.ymland run the following command:
docker rmi <image_name:tag>To get the latest image after updates:
docker compose pull
docker compose upconda(version 23.9.0 or higher)conda-lock(version 3.0.4 or higher)jupyterlab(version 4.0.0 or higher)nb_conda_kernels(version 2.3.1 or higher)- Python and packages listed in
environment.yml
If you modify environment.yml:
- Regenerate the conda-lock file:
conda-lock lock -f environment.yml- Push the changes into your own branch and GitHub Actions workflow will automatically build and push the new image:
git add conda-lock.yml
git commit -m "Update conda-lock.yml file"
git push origin <branch_name>-
Update the `docker-compose.yml file on your branch to use the new image (specifically, update the tag).
-
Send a pull request to merge the changes into the main branch.
Unit tests are run using the pytest command in the root of the project.
# Run all tests
pytest -v
# Run specific test file
pytest tests/test_download_data.py -vThis project utilizes the Student Performance Dataset from UCI Machine Learning Repository, which is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. This allows for the sharing and adaptation of the datasets for any purpose, provided that the appropriate credit is given.
The Student Grade Predictor report contained herein is licensed under the Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) License. See the LICENSE.md file for more information. If re-using/re-mixing please provide attribution and link to this repository.
The software code contained within this repository is licensed under the MIT license. See the LICENSE.md file for more information.
Cortez, P. (2008). Student Performance [Dataset]. UCI Machine Learning Repository. https://doi.org/10.24432/C5TG7T.
Cortez, P. & Silva, A. (2008). Using data mining to predict secondary school student performance. EUROSIS. https://doi.org/10.24432/C5TG7T.
Ma, Y., Liu, B., Wong, C., Yu, P., & Lee, S. (2000). Targeting the right students using data mining. Proceedings of the sixth ACM SIGKDD international conference on Knowledge discovery and data mining, 457-464. https://doi.org/10.1145/347090.347184.
Pritchard, M. & Wilson, G. (2003). Using Emotional and Social Factors to Predict Student Success. Journal of College Student Development, 44, 18-28. https://doi.org/10.1353/csd.2003.0008.
Johora, F. T., Hasan, M. N., Rajbongshi, A., Ashrafuzzaman, M., & Akter, F. (2025). An explainable AI-based approach for predicting undergraduate students academic performance. Array, 26, 100384. https://doi.org/10.1016/j.array.2025.100384.
This project is a demonstration for educational purposes, and the structure and workflow was adapted from the "Breast Cancer Predictor" project by Tiffany Timbers, Melissa Lee, Joel Ostblom & Weilin Han.