An Experimental Study on Hyper-parameter Optimization for Breast Cancer Risk Prediction using IBM Snap Machine Learning Techniques

Worldwide approximately 1 in 6 women will succumb to Breast Cancer (BC), making it one of the most lethal illnesses in the female population. To avoid and receive adequate clinical diagnostic therapy, it is imperative that early detection be made possible. Medical applications rely heavily on Machine Learning (ML) for risk prediction, therapy recommendation, and critical decision assistance. With the Wisconsin BC diagnosis dataset from UCI (University of California Irvine) repository, we have evaluated two IBM Watson’s ML algorithms such as Snap Logistic Regression (LR) and Snap SVM Classifier. The algorithm’s goal is to identify the diseased cells as Benign (B) or Malignant (M) that is either non-cancerous or cancerous. Performance metrics such as Accuracy, AROC (Asymptotic Receiver Operating Characteristic), Precision, Recall, F1, Average Precision, and Log Loss are computed to measure the model outperformed the rest of the algorithm in all other measures, with a cross validation score of 98.2 percent.