This project implements and compares multiple classical machine learning models and a neural network for breast cancer diagnosis using the Wisconsin Breast Cancer Diagnostic dataset.
The goal is to benchmark different models, evaluate their predictive performance, and provide a clean, reproducible ML pipeline suitable for academic and applied use.
- Source: Wisconsin Breast Cancer Diagnostic Dataset
- Task: Binary classification (Benign vs Malignant)
- Target variable:
diagnosis- Malignant → 1
- Benign → 0
Non-informative identifiers (e.g., id) are removed during preprocessing.
- Random Forest
- Gradient Boosting
- Neural Network (Keras / TensorFlow)
- Soft Voting Classifier (ensemble)
- Data loading and cleaning
- Feature scaling using StandardScaler
- Train / test split (80/20)
- Model training
- Performance evaluation
- Visualization and model persistence
- Accuracy
- Precision, Recall, F1-score
- Confusion Matrix
- Feature importance plots
- Confusion matrix
- Trained models saved for reuse
- Scaler for reproducibility
All outputs are stored in the results/ and models/ directories.
pip install -r requirements.txt
python main.py