MTHO-ES2A-1DLTM: MULTI-OBJECTIVE TRAP HUNT ATTENTION ENABLED DEEP LEARNER IN ARRHYTHMIA DETECTION AND CLASSIFICATION

Context: Arrhythmia detection and classification play a significant role in cardiovascular health monitoring, requiring reliable and efficient diagnostic tools. There exist several methods in the detection of arrhythmia that encounter significant challenges, such as high false positive rates and limited sensitivity to rare arrhythmias, impacting diagnostic accuracy and patient care. Objective: To address these issues, the research introduces a Multi-Objective Trap Hunt Algorithm Optimized Enhanced Spatial-Shuffle Attention–enabled One-Dimensional Convolutional Neural Network coupled with Long Short-term Memory (MTHO-ES2A-1DLTM) model, which is the combination of 1D Convolutional Neural Networks (1DCNN) and Long Short-Term Memory (LSTM) networks that works with the photoplethysmography (PPG) signal. Methods: In the proposed model, the 1DLTM leverages feature extraction from PPG signals and further captures temporal dependencies that are crucial for arrhythmia classification. Additionally, the model integrates optimization strategies inspired by the characteristics of antlion and sea lion behaviors, and also the hybrid attention mechanism of ES2A is included to yield remarkable features. Results: The proposed model achieved Precision, Recall, Accuracy, and [Formula: see text]-score of 97.97%, 98.62%, 98.63%, and 98.3%, respectively. Significance: Moreover, the proposed model offers several advantages over traditional approaches, including improved accuracy in distinguishing between different arrhythmia types, reduced false positives, and enhanced sensitivity to rare and complex arrhythmias.