This project implements a machine learning pipeline for breast cancer prediction using DNA methylation data. The pipeline includes data preprocessing, feature selection using Boruta, and model training with XGBoost.
The project includes two models: the Orginal Model which was done based on the paper, and an Improved Model that enhances some of the metrics of the Orignal Model.
Setup
Create a virtual environment:
python -m venv venv
Install dependencies:
pip install -r requirements.txt
Usage
Original Model Pipeline
python run_pipeline.py
Improved Model Pipeline
python run_improved_model.py
Model Results
Original Model Performance
Validation Metrics
ROC AUC: 0.9261
Accuracy: 0.9189
Sensitivity: 0.2500
Specificity: 1.0000
Test Metrics
ROC AUC: 0.8056
Accuracy: 0.8987
Sensitivity: 0.0588
Specificity: 1.0000
Improved Model Performance
Key Improvements
Class Imbalance Handling:
Implemented class weights (scale_pos_weight = 8.3846)
