This project applies logistic regression to classify breast cancer tumors as malignant or benign using the Breast Cancer Wisconsin dataset.
The main goals are to:
- Explore and preprocess real-world medical data
- Build and evaluate a logistic regression classifier
- Interpret coefficients to understand key predictive features
- Discuss limitations and ethical considerations of applying ML in healthcare
- Source: [Kaggle – Breast Cancer Wisconsin (Diagnostic)]
- Size: 569 rows, 31 numeric features
- Target: Diagnosis (Malignant = 1, Benign = 0)
- Python (pandas, numpy, matplotlib, seaborn)
- scikit-learn (Logistic Regression, metrics, preprocessing)
- Jupyter Notebook for documentation and analysis
- Exploratory Data Analysis (EDA): Visualize class balance and feature distributions.
- Preprocessing: Standardize features to make coefficients comparable and encode categorical features.
- Modeling: Train a logistic regression classifier.
- Evaluation: Assess accuracy, precision, recall, F1, confusion matrix, and ROC-AUC.
- Interpretation: Analyze model coefficients.
- Discussion: Highlight ethical implications and limitations.
- Logistic regression achieved near-perfect ROC-AUC on this dataset due to its clean separability.
- Features related to tumor size, texture and shape (e.g., mean radius, concavity) were the strongest predictors.
- The model is not intended for clinical use but demonstrates how ML methods can support healthcare analytics.
- Predictive models in healthcare require rigorous testing, validation, and expert oversight.
- False negatives (missed malignant cases) are especially harmful and must be carefully addressed.
- This project is a proof of concept for learning purposes only.
- Logistic regression is both interpretable and effective for binary classification.
- Standardizing features is critical for fair coefficient comparison.
- Even when performance is high, transparency and ethics are essential in healthcare ML applications.