Optimizing the Classification Performance by Fine-Tuning the Machine Learning Hyperparameters and Utilizing PCA and RFE Feature Selection Methods

Breast cancer is among the most common and potentially fatal cancers, especially among women. Breast cancer usually has no obvious early signs, and doctors sometimes have trouble differentiating benign and malignant tumors. Thus, with the help of appropriate Machine Learning (ML) techniques, breast cancer patients can be classified more accurately and help doctors make diagnosis decisions. The research gaps in this topic are that there is no standard ML hyperparameter setting and feature selection method used to improve the ML algorithm performance in breast cancer classification. This paper aims to identify the best hyperparameters settings, identify the best feature selection methods and compare them with past research papers. The experiments conducted in this paper can be divided into two: the first experiment is to identify the best hyperparameter settings, and the second is to identify the best feature selection methods. In this paper, several algorithms are used that includes Logistic Regression (LR), Decision Tree (DT), k-Nearest Neighbors (k-NN), Support Vector Machine (SVM), Random Forest (RF), AdaBoost, Hard Voting, and Artificial Neural Network (ANN) to model the Wisconsin Diagnostic Breast Cancer (WDBC) dataset. Meanwhile, the feature selection method uses Principal Component Analysis (PCA) and Recursive Feature Elimination (RFE). For comparison with past research, research papers that have used the same dataset and algorithms will be selected to ensure the comparison is fair and has no bias. Based on the results obtained in this paper, Adaboost, combined with the optimized decision tree, is found to be the best algorithm due to high accuracy and sensitivity compared to the other algorithms, which are 98.2% and 99.6%.