Student Habits and Exam Performance

Machine-learning study of the relationship between student routines, background factors, and exam performance.

The analysis compares linear and regularized regression, random forest, gradient boosting, support vector regression, and multilayer perceptron models. PCA, K-means, and Gaussian-mixture clustering are used to examine lower-dimensional structure and student profiles.

Results

The primary dataset contains 1,000 student records. A tuned radial-basis SVR produced the strongest verified held-out result.

Metric	Result
Cross-validation R²	0.894
Test R²	0.890
Training R²	0.918

Study time, attendance, sleep, mental-health indicators, and screen-time behavior were among the factors examined. The analysis is predictive and does not establish causal effects.

Repository contents

• notebooks/analysis.ipynb - exploratory analysis, modeling, validation, PCA, and clustering
• data/student_habits_performance.csv - primary dataset
• data/real_student_data.xlsx - small exploratory validation dataset
• reports/final-report.pdf - final project report

Run

python -m pip install -r requirements.txt

Open notebooks/analysis.ipynb and run the cells in order.

Limitations

• The main dataset is observational and may contain self-reported variables.
• The small external workbook is exploratory and is not directly comparable with the main test set.
• Performance may change in a different school or student population.

Attribution

Individual academic project completed for Georgia Tech OMS Analytics.