Integrating Feature Selection and Machine Learning Boosting for Accurate Breast Cancer Prediction

Abstract

Breast cancer is a prevalent and devastating disease and remains a major contributor to cancer-related mortality among women worldwide. The increasing incidence and fatality rates are often associated with changes in lifestyle and the influence of environmental factors. In response to these alarming trends, the development and deployment of automated breast cancer diagnostic systems have become increasingly important in modern healthcare. This study investigates the performance of several boosting algorithms - CatBoost, LightGBM, XGBoost, AdaBoost, and Gradient Boosting - for breast cancer prediction using the Wisconsin Diagnostic Breast Cancer (WDBC) dataset. The dataset is publicly available on Kaggle and consists of 569 instances, including 357 benign and 212 malignant cases. The proposed framework encompasses data preprocessing, feature selection, and classification stages. Model performance was evaluated using multiple metrics to ensure robust analysis and objective assessment. The experimental results demonstrate that LightGBM outperformed the other models, highlighting the effectiveness of boosting-based approaches for breast cancer diagnosis and emphasizing the potential of these techniques for further advancements in oncology research.