Abstract

The quality of electrocardiogram (ECG) signals plays a crucial role in the performance of deep neural network (DNN) models on cardiac arrhythmia classification. The performance of deep neural network (DNN) models on cardiac arrhythmia classification is highly influenced by the quality of electrocardiogram (ECG) signals. This paper presents a novel optimization-based preprocessing framework CS-TS-DNN that combines Cubic Spline interpolation with Tabu Search (TS) metaheuristic for the automatic selection of optimal spline control points to represent the ECG signal. The proposed model can be used to optimize the data adaptively for improved classification accuracy and signal morphology preservation without employing heuristic approaches to pre-processing. The proposed approach has two stages. ECG signals are normalized and class imbalance is addressed using the Synthetic Minority Oversampling Technique (SMOTE). Second, Cubic Spline interpolation is applied using both manually selected and Tabu Search-optimized spline control points. To search the space of spline-lengths for the optimal configuration, Tabu Search has used 20 iterations. The proposed model has been tested on the Arrhythmia dataset from both INCART and MIT-BIH. The experimental results showed that the TS-based optimization method outperformed the manual spline selection. The proposed model gave an optimal spline length of 19 and achieved 98.75% accuracy on INCART, whereas manual selection resulted in 98.27% accuracy. The second validation on MIT-BIH has 97.75% accuracy with an optimized spline length of 162. The results showed that adaptive preprocessing optimization enhances the quality of representation, classification accuracy, robustness, and generalizability of the ECG across different datasets.