International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 04 | Apr 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3360 Neural Network-Based Automatic Classification of ECG Signals with Wavelet Statistical Characteristics Arjun Choudhary 1 , Dr. Kalpna Sharma 2 & Dr. Prakash Choudhary 3 1 Research Scholar, Computer Science Department, Bhagwant University, Ajmer, Rajasthan-305001, India 2 HOD & Assistant Professor, Computer Science and Engineering Department, Bhagwant University, Ajmer, Rajasthan-305001, India 3 Assistant Professor, Computer Science and Engineering Department, National Institute of Technology Hamirpur, HP-177005, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - Cardiac abnormalities are the most common threat to human life. An electrocardiogram is the most common way to examine a heart abnormality. We present automatic detection of two types of ECG signals with statistical wavelet features using a Multilayer Perception Neural Network as the classifier. The database used for the heart abnormality detection is the MIT-BIH arrhythmia database. The Butterworth and Chebyshev Type-II filters have been introduced for de-noising the signal. The wavelet features have been extracted from the preprocessed ECG signal by using DWT (discrete wavelet transform) and 3600 samples have been taken from each signal and split into frames. The total number of samples of the signal is split into 4 windows, and each window contains 900 samples. DWT is applied in each frame or window to get wavelet coefficients which determine the characteristics of the signal. This wavelet coefficient is the input feature of the classifier for training and testing the model, which gives up to 100% accuracy for normal cases and 90% abnormality detection. This has been achieved. Key Words: MIT-BIH Arrhythmia, DWT, ECG, LDA, MLP, Chebyshev Type-II, Neural Network, Perceptron. 1. INTRODUCTION There are mainly three sorts of components within the ECG signals. Each wave contains different information, which has includes amplitudes, durations, and morphology. Then High blood pressure, cholesterol, smoking, being overweight, etc. are the various causes that increase the general risk of heart disorder. During long-term monitoring, an automatic analysis of the ECG signal is vital to classify the various diseases of the heart. Manually analysing an oversized amount of information could be a very time-consuming task for doctors and analysts. Hence, there's a necessity for computational methods and machine learning techniques for the classification of the ECG signal. ECG analysis tools require knowledge of the location and morphology of the varied segments in the ECG recordings [1], [2]. Karpagechilvi et al. [3] proposed a sentimental analysis method where it's necessary to extract vital information from the ECG signal to detect new features for his use as an input within the artificial neural network to classify the ECG signal. In past studies, many researchers have worked with the ECG signal to detect the heart disorder. Several algorithms are been developed for the classification of ECG signals. Stalin Subbiah et al. [4] proposed a method for preprocessing to cancel the noise using Gaussian filters, median filters, FIR-filters, and Butterworth filters. These are used for feature extraction, wavelet transformation, and QRS component features are used as a classifier input to spot the conventional and abnormal heartbeat. Eduardo Joseda S. Luz et al. [5] performed research by which a way is proposed which uses a 10 sec ECG signal for normal and arrhythmia or abnormal ECG classification. The database has been taken from the MIT-BIH normal sinus database and supraventricular arrhythmia database. Sharma and Bhardwaj et al. [6] proposed the model to train the neural network. The Levenberg-Marquardt function is used with 100% accuracy for the normal Sinus Database. Ayub, J.P. Saini et al. [7]. Research performed by him uses the ANFIS (Adaptive Neuro-Fuzzy Interface System) model to identify normal and abnormal ECG signals. The MIT-BIH normal sinus and MIT-BIH supraventricular databases are used for training and testing the neural network. A feed- forward and back-propagation algorithm are accustomed minimise the errors, and a trapezoidal member function is employed as an input and output. Mondal, S., Choudhary, P., et al. [8] used this MIT-BIH normal sinus database. The records from the MIT-BIH Arrhythmias and Apnea ECG databases from Physionet are used for training and testing our neural network based classifier. From which 90% healthy and 100% abnormal records are detected within the MIT-BIH Arrhythmias database with an overall accuracy of 94.44%. Within the Apnea-ECG database, 96% of normal and 95.6% of abnormal ECG signals are detected, achieving a 95.7% classification rate. From this MIT-BIH normal sinus database, 18 samples are taken and 61 samples are taken for abnormalities to train and test the model. The proposed model gives an accuracy of 100% for normal and 91% for abnormal.