Minimum feature set based opitmally tuned adaptive classification algorithm for arrhythmia detection

This paper presents an optimally tuned adaptive classification algorithm called the PORS (PCA ReliefF-based Optimized SVM) method for arrhythmia detection using a minimum feature set. Arrhythmia is an abnormality in the heart that can be detected by sensing the electrocardiogram signal (ECG) and analyzed with respect to a definite set of features. The number and nature of features required for arrhythmia classification will vary for each type of arrhythmia. Selecting the appropriate features and classifying the arrhythmia precisely with minimal computational complexity is a challenging task. The proposed PORS method consists of an optimally tuned support vector machine with the minimum number of inputs derived from principal component analysis along with ReliefF ranking. Initially, we consider nine features, which are minimized to seven by PCA in the first level. They are further reduced to 3 using the ReliefF ranking. We use SVM for accurate arrhythmia classification and further enhance its accuracy, sensitivity, specificity, and predictive values with meta-heuristic optimization techniques. Through experimental results, we show that adaptively tuned SVM can be used for both single and multiple classifications in a single run. We also prove through comparison of results that our PORS method performs better than popular classification methods such as Naïve Bayes, k-Nearest Neighbor, and conventional SVM. The major contribution in our proposed work is twofold: we minimize the feature set using a combination of PCA and ReliefF algorithms, which drastically reduce the inputs without affecting the performance. Secondly, we use the optimally tuned adaptive SVM method for classification, which greatly improves the accuracy.