Deep Learning Driven Heart Disease Prediction using ECG Signal Classification

The non-stationary motivations of the ECG structures are significant readings to estimate the pace and rhythm of the heartbeat and therefore to estimate the health of the heart. This paper proposes a robust method for automatic ECG classification, segmenting data into two categories: variations in normal and pathological cardiovascular patterns. More specifically, the collected ECG signal is normalized, filtered to eliminate noise interference. Hence morphological attributes are formed to represent the feature of the classes to the best of their ability. ACO is incorporated into segmentation and feature selection to reduce feature vectors of abnormalities. The classification of ECG signals is made through using a Bi-Long Term Short Term Memory (Bi-LSTM) neural network because it is a network proven for sequence prediction and it is capable to capture long temporal dependence in the data. Evaluation criteria show that, relative to other proposed classifications of ECG, the presented approach reaches 90% classification accuracy. It has been established that the integration of ACO with Bi-LSTM provides improved classification of ECGs thereby making it a useful tool in clinical diagnosis and therapy monitoring.