Feature Extraction and Interpretation of ECG Signals

Abstract

newline Healthcare is highly dependent on the information provided by individual physiological signals. It is possible to learn a lot about a person emotional, behavioral, and cardiovascular health by observing their ECG cyclic behavior. This thesis developed two optimization techniques for feature selection and deep learning for heartbeat classification to identify cardiac arrhythmias in ECG signals. newlineThe first proposed work Morphological feature and wavelet coefficient based features were extracted from the recorded ECG signals. The Improved Monarch Butterfly optimization approach is used to reduce the dimensionality of the feature vector. Using convolution neural networks, these characteristics are used to categorize signals. The experimental findings of this method shows 99.49 accuracy, 99.58 sensitivity, and 98.83 specificity which is comparable to earlier approaches, which may aid in diagnosing an arrhythmia by a doctor. newlineThe wrapper feature selection technique incorporates a Pigeon inspired optimizer in the second proposed work. The Deep Neural Network is utilized to identify ECG data using a modified Pigeon Inspired Optimizer. MPIO new blood pigeons have improved the algorithm accuracy. ECG signals may be classified using morphological characteristics, wavelet transform coefficients, and R R interval dynamic features. Optimizing the feature plays a vital role in constructing a machine learning model because irrelevant data features impair model accuracy and increase model training time. MPIO is used for feature optimization after feature extraction. The DNN classifier is used to categorize ECG data using optimal features. The suggested technique achieves an accuracy of 99.10, 98.90 specificity, and 98.50 sensitivity. Additionally, compared to other current methods, our feature selection approach yielded more significant results. newline