Feature Selection Techniques on Breast Cancer Classification Using Fine Needle Aspiration Features: A Comparative Study

Breast cancer remains a prevalent invasive cancer in women as it ranks as the second leading cause of cancer-related death among women. It poses a significant global medical challenge due to its substantial increase in cases over the last decade. Early detection of breast cancer is vital; hence the development of computer-aided diagnosis (CAD) systems is crucial in assisting pathologists to accurately interpret and diagnose the tumor. Feature selection plays a significant role in CAD as it involves choosing the most relevant and informative features from the original dataset to improve the performance of the system. Thus, this study focuses on evaluating various feature selection methods on fine needle aspiration (FNA) features which are adapted from Wisconsin Diagnostic Breast Cancer (WDBC) dataset from UCI Repository. The analysis involved five feature selection techniques; Information Gain (InfoGain), Correlation Feature Selection (CFS), Fast-Correlation Based Filter (FCBF), Consistency and Relief-F with three different machine learning classifiers including Logistic Regression (LR), Support Vector Machine (SVM) and Random Forest (RF) with 10-fold cross-validations. Based on the experimental outcomes, it was observed that FCBF with LR classifier surpassed other FS techniques (ACC = 0.9718 and AUC = 0.993) with 7 features. On the other hand, Relief-F outshined other FS with both classifiers of SVM (ACC = 0.9772 and AUC = 0.971) and RF (ACC = 0.9684 and AUC = 0.991). This study validated that the Relief-F technique exhibited supremacy over other FS techniques. However, the task of identifying important features from high-dimensional data remains a significant obstacle in intelligent diagnosis. Henceforth, it is essential to dedicate further efforts to the development of CAD systems using efficient feature selection techniques to maximize the performance and effectiveness of diagnostic models.