An AIoT Enabled Multi-Level Decision Support System for Remote Arrhythmia Analysis Using Efficient Wavelet Transform

In this work, an artificial intelligence of things (AIoT) framework for remote cardiac health monitoring includes wearable sensor devices in the perception layer and an AI model in the cloud server’s application layer with an IoT cloud interface, is proposed. The AI model works as a multi-level ECG feature detector and an accuracy predictor in the proposed work. Three wavelet transforms, namely Discrete Wavelet Transform (DWT), Dual-Tree Complex Wavelet Transform (DTCWT), and Maximal Overlap Discrete Wavelet Transform (MODWT), are experimentally tested and analyzed for the AI model. The wavelet functions like Daubechies and Symlet are compared for maximum R-peak detection. Various classifiers like FNN and SVM are employed to evaluate the performance of the proposed model on parameters like accuracy, recall, F1Score, and ROCs. Using the ECG’s unique features, an accuracy of 99.8% for abnormality detection has been achieved by SVM. Using MODWT extracted features, SVM outperforms FNN having 97.4% accuracy for the type of abnormality check. FNN also achieved 97.5% accuracy for abnormality detection. The benchmark PhysioNet datasets are validated for two class and four class classifications comprising big dataset management.