Efficient myocardial ischemia classifier based on statistical features with random weight settings

The novelty of this work is extracting the multiple statistical features from denoised ECG beat segment for an ANN classifier which is trained and tested by considering 10 different values of random weights and biases for optimum choice of classifier architecture. The proposed MLP neural network receives the statistical features extracted from preprocessed ECG beat segment and trained with Levenberg-Marquardt algorithm. To demonstrate the efficacy of the ANN classifier, training and testing datasets are chosen from European ST-T datasets of physiobank database. The performance of ANN model is compared with K-nearest neighbor (KNN) and support vector machine (SVM) classifiers. The experimental results confirmed that the ANN model with 12 hidden neurons outperformed with overall classification accuracy of 92.85 %.