Classification of Patients with Breast Cancer using Neighbourhood Component Analysis and Supervised Machine Learning Techniques

摘要

Breast cancer is considered one of the leading causes of death among women. In morocco, the ministry of health reports over 40.000 new cases each year. When lifestyle can be a preventive pattern, early detection remains a factor of a huge impact on the mortality of the diseases. Machine learning (ML) algorithms offer an alternative to breast cancer standard techniques of prediction, or at least can assist radiologists in their reasoning flow and thus saving many females and some males from breast cancer biopsy. The present study represents a benchmarking of different ML models. The research applies and compares four machine learning algorithms (kNN, decision tree, Binary SVM, and Adaboost) to predict whether a patient has a malignant or a benign tumor. The machine learning techniques have been trained then tested on the Breast Cancer Wisconsin dataset. The datasets features are fed into feature selection model with Neighbourhood Components Analysis (NCA) to reduce the number of features and therefore decrease the complexity of the model. The predictive accuracy reached a 99.12% for the kNN model, the best predictive specificity obtained was 9S.S6% for the Binary SVM model and the highest predictive sensitivity obtained was up to one for both kNN and Adaboost models.