ANALYSIS AND CLASSIFICATION OF HUMAN PHYSIOLOGICAL SIGNALS

Abstract

The results of the proposed scheme with performance measures like accuracy, specificity, sensitivity, precision, Jaccard coefficient, and dice co-efficient are compared with the existing techniques for proving efficacy. For Arrhythmia detection, normal and arrhythmia signals waveforms are analysed and their patterns are represented. In this, the automatic method of arrhythmia classification is performed using the proposed Bat-Rider Optimization algorithm-based deep convolutional neural networks (BaROA-based DCNN). The features are fed to the arrhythmia classification module, which classifies the patient as either affected with arrhythmia or normal. The classifier Deep CNN yields an accurate classification and it is an automatic way of classification. The experimentation is performed using the MIT-BIH Arrhythmia Database and the analysis is performed based on the evaluation metrics. The proposed research work deals with Epileptic seizure detection using the machine learning methods based on EEG Signals. The abnormal electrical disturbances are normally termed as seizure. The improvement of quality treatment is the most essential for epileptic patients. The limitations of existing classification techniques are unknown network duration, having a distributed memory and fault tolerance. So, we need a new framework for detecting cardiac abnormalities and epileptic seizure classification. This work proposed the novel pre-processing technique named as Enhanced Curvelet Transform (ECT) and new hybrid model for feature extraction in epileptic seizure detection. The proposed hybrid model combines the methods of MGT (Modified Graph Theory), NPT (Novel Pattern Transformation), and GLCM feature extraction. Finally, the robust classification technique is achieved by using PCA based Random forest classification. This proposed classification technique had the advantages of high accuracy, minimum overfitting, minimum information loss, and insensitivity to noise. The results of proposed method are analysed by various performance analysis. Another enhancement in this work is proposed to classify the epileptic seizure using neural network based on their frequency waveforms. In this work, the Modified Blackman Bandpass Filter (MBBF) is used for removing the artifacts from the EEG signal. Then the time domain and frequency domain features are retrieved and optimized using Greedy Particle Swarm Optimization. Finally, the classification of seizure is achieved by using the Convolutional Neural Network. The CNN classification utilized to classify the ictal, inter-ictal and healthy classes. Based on the experimental results, it can be observed that the proposed method exhibits enhanced efficiency and accuracy in comparison to other existing techniques.