Evaluasi Recursive Feature Elimination Untuk Klasifikasi Kanker Payudara Menggunakan Berbagai Algoritma Machine Learning

Abstract

Early detection of breast cancer requires classification models that are not only accurate but also efficient and interpretable. This study evaluates the effect of Recursive Feature Elimination (RFE) on the performance of several machine learning algorithms for breast cancer classification. The dataset used is the Wisconsin Diagnostic Breast Cancer (WDBC) dataset from the UCI Machine Learning Repository, consisting of 569 samples and 30 numerical features. The research stages include data preprocessing, removal of non-informative attributes, feature standardization using StandardScaler, train-test splitting with an 80:20 ratio, feature selection using Logistic Regression-based RFE, and training and testing of 11 classification algorithms. Model performance was evaluated using accuracy, precision, recall, F1-score, confusion matrix, and Receiver Operating Characteristic (ROC) curve. The results show that before feature selection, Support Vector Machine, Logistic Regression, and Voting Classifier achieved the highest accuracy of 98.25%. After applying RFE, the accuracy of these models decreased slightly to 97.37%, while the number of features was reduced from 30 to 15. Several algorithms, including Nearest Centroid, Naïve Bayes, and AdaBoost, showed improved accuracy after RFE. These findings indicate that RFE does not always improve the best model accuracy, but it can produce a more compact, efficient, and interpretable classification model.