<![CDATA[Timely identification of breast cancer recurrence is closely associated with patient survival and the effectiveness of treatment. Inaccurate detection can contribute to greater disease severity, higher treatment costs, longer recovery, and reduced quality of care. For Machine Learning (ML)-based decision-support systems, two important challenges are the unequal distribution of medical-data classes and the large number of features, both of which may affect model accuracy and computational efficiency. This study evaluates an approach that combines feature selection with class-imbalance handling to improve breast cancer detection performance. Information Gain (IG), Gain Ratio (GR), Gini Decrease (GD), and Relief-F are used to rank features according to their weights, while the Synthetic Minority Over-Sampling Technique (SMOTE) is applied to improve representation of the underrepresented class. Seven ML classifiers, namely k-Nearest Neighbor (k-NN), Tree, Support Vector Machine (SVM), Naive Bayes, AdaBoost, Random Forest (RF), and Neural Network (NN), are tested and assessed using confusion-matrix-based accuracy, precision, recall, and computational time. The experimental results indicate that incorporating class-imbalance handling improves the predictive performance of the ML algorithms. Among the evaluated combinations, Information Gain with Random Forest (IG+RF) provides the optimal result in this case. These findings highlight the value of integrating class-balancing and feature-selection procedures when developing machine-learning systems for breast cancer detection.]]>
Copyrights © 2026
