Feature Selection Strategies for Efficient Breast Cancer Classification

This project explores how selecting the right subset of features can improve the efficiency, accuracy, and interpretability of machine learning models for breast cancer diagnosis. Using the Breast Cancer Wisconsin (Diagnostic) dataset, the study analyzes how different groups of input features—such as mean values, standard errors, and worst-case measurements—affect classification performance when using a Random Forest classifier. The core of the project is a custom Python framework that allows flexible experimentation with various feature combinations. It enables users to systematically train and evaluate models on different subsets of the dataset’s 30 features, track accuracy, and visualize results. Through two carefully designed experiments, the study demonstrates that worst-case measurements and a small set of top-ranked morphological features (like radius, texture, perimeter, and area) can achieve near-maximum classification accuracy—often outperforming models trained on all 30 features. These findings highlight the potential of dimensionality reduction for building lightweight, interpretable diagnostic models in medical settings. All code and results are provided in a public GitHub repository to support reproducibility and future exploration.