Authors : Mohammad-Ashraf Ottom and Walaa Alshorman
Abstract: Diagnosis of heart disease is a very critical and challenging task in medical science field. Healthcare providers are collecting a massive amount data including heart disease data which unfortunately contains relevant, irrelevant and redundant features. Selecting optimal features set is a critical and important process for such high dimensional data analysis. Data mining and features selection techniques can assist in decision making and for better diagnostics of many diseases, however, traditional features selection techniques provide a limited contribution to classification. This research proposes a new approach for features selection. The new approach suggests collaboration between well-known features selection techniques by accumulating features rank for all selected features selection techniques, then, elects features which produce the highest rank. The optimal features are used to form a subset of dataset and features with lower rank will be eliminated. The full dataset and new dataset will be evaluated on five well known machine learning algorithms to evaluate the result of the proposed technique according to accuracy, recall, precision and f-measure. The proposed approach shown better prediction rate based where kNN recorded the best enhancement rate in accuracy (+3.7%), precision (+8.1%), recall and (+0.3.5%) f-measure (+0.5.7%).
Mohammad-Ashraf Ottom and Walaa Alshorman, 2019. Heart Diseases Prediction Using Accumulated Rank Features Selection Technique. Journal of Engineering and Applied Sciences, 14: 2249-2257.