Diagnosis of Breast Cancer and Diabetes using Hybrid Feature Selection Method

Diagnosis of diseases at an early stage is a crucial task in the medical field. A hybrid machine learning framework is presented for the diagnosis of breast cancer and diabetes using efficient feature selection and classification technique. This research identifies significant risk factors related to both chronic disease datasets by applying different feature selection techniques and hybridization of ReliefF Feature Ranking with Principal Component Analysis (PCA) method. To evaluate the effectiveness of the presented feature selection method, k-nearest neighbor method for classification is used. The hybridization enhances the accuracy of the classifier with the proposed feature selection technique for both chronic disease datasets. The performance of the presented hybrid framework is found to be best in comparison to five other techniques - Correlation Based feature Selection (CBS), Fast Correlation Based Feature Selection (FCBF), Mutual Information Based Feature Selection (MIFS), MODTree Filtering Approach and ReliefF Feature Selection. Moreover, the proposed ReliefF-PCA method eliminates 25% and 33.3% of irrelevant features for diabetes and breast cancer dataset respectively.