Chaotic-based divide-and-conquer feature selection method and its application in cardiac arrhythmia classification

Feature selection by removing redundant and noisy features is one of the crucial steps in the classification problem. This paper presents a novel chaotic-based divide-and-conquer (CDC) algorithm to select optimal features from an available feature set (the UCI Arrhythmia Dataset). We then employed it for a quick and automatic heart function examination which is essential for monitoring the heart functionality of risky patients. The method begins with chaos numbers to select several features as cluster-heads. We used chaos sequences to escape from the dependency on initial values and getting stuck in local optima. Then, it assigns each feature to a group of cluster-heads and finally selects a representative from each group. The proposed method resulted in performance rates of 88.21%, 89.41%, 87.64%, and 86.54% in terms of accuracy, sensitivity, specificity, and F-measure, respectively. Since this method removes the redundant or improper features of the dataset without any data loss, it approximately needs 0.6 seconds to diagnose and classify cardiac arrhythmias. It is highly time-effective compared to the current state-of-the-art approaches.