Feature Selection and Data Imbalance Handling for Breast Cancer Classification Using the Wisconsin Diagnostic Breast Cancer (WDBC) Dataset

Abstract

Data quality is one of the factors that influence the performance of machine learning algorithms in classification tasks. Issues such as a high number of features and class distribution imbalance can lead to a decrease in accuracy and the model's ability to recognize data patterns. This study aims to analyze the impact of feature selection methods and data imbalance handling techniques on the performance of a breast cancer classification model using the Wisconsin Diagnostic Breast Cancer (WDBC) dataset. The feature selection methods used include Principal Component Analysis (PCA), Chi-Square, SelectKBest, and Recursive Feature Elimination (RFE), while the data balancing techniques applied are SMOTE and UnderSampling. All combinations of these methods were tested using the Logistic Regression, Decision Tree, and Random Forest algorithms. The results of the study show that feature selection methods and data imbalance handling techniques influence the performance of classification models. Numerically, the combination of PCA and undersampling in the logistic regression algorithm yielded the highest performance, with an accuracy of 97%, precision of 95%, recall of 98%, and an F1-score of 96%. However, based on an evaluation of the learning curve, generalization ability, and model interpretability, the combination of Chi-Square and undersampling with the Logistic Regression algorithm was selected as the best model. This combination yielded an accuracy of 95%, precision of 95%, recall of 90%, and an F1-score of 93%, and demonstrated a "good fit" with strong generalization ability. The research results prove that selecting the appropriate preprocessing method can improve the quality and reliability of breast cancer classification models.