Integrating Multiscale Mapper Features with Machine Learning Techniques to Improve Breast Cancer Classification Accuracy
Rabelani Netshifhire, Maria Vivien Visaya · 2024
Breast cancer classification is critical for early detection and treatment planning. The complexity of breast cancer data poses feature engineering challenges for conventional machine learning (ML) methods, which often fail to capture these complexities.We employ the Mapper algorithm, a topological data analysis (TDA) tool, for feature engineering. Two variants of Mapper algorithm are utilized, namely Singlescale Mapper (SM) and Multiscale Mapper (MM). SM is constrained by scale sensitivity, dimensionality limitations, and parameter tuning challenges. MM enhance SM by capturing topological features at multiple resolutions, offering a robust hierarchical view of data.Using the Wisconsin Breast Cancer Diagnostic (WBCD) dataset, three feature sets namely conventional, SM-derived, and MM-derived features are evaluated across various selected ML models to assess their potential to enhance classification accuracy.Results demonstrated that MM-derived features integrated with ML models improved breast cancer classification accuracy, outperforming both conventional and SM-derived features. This findings highlight the potential of TDA’s MM as a feature engineering tool to improve breast cancer classification.