Advancements in remote healthcare monitoring: A comprehensive system for improved health management

Abstract

The research focuses on improving the prediction of heart illness using ECG sensor data by applying sophisticated machine-learning algorithms. Preprocessing of ECG signals is important for extracting the cardiac waveform and improving the data quality. In this regard, baseline drift and low-frequency noise have been removed by applying a Gaussian filter. This preprocessing enables a better investigation of the components of the ECG, including the T-wave, QRS complex, and P-wave. In predicting cardiac disease, we use two variants of neural networks: Multilayer Perceptron and Gated Recurrent Units. The MLP is one variant of feedforward artificial neural networks that process pre-processed ECG data through a many-layered network. ReLU activation functions introduce non-linearity and map raw data into higher-dimensional representations that capture the essential properties. GRU is a variant of the RNN that reduces the problem of the vanishing gradient, which affects conventional RNNs. Therefore, it uses update and reset gates to better handle sequential data like ECG data. From our findings, GRU could perform better than MLP in most performance parameters, such as accuracy, precision, recall, F1 score, and AUC-ROC. The superior performance of GRU is attributed to the fact that it can decode complex temporal patterns related to heart disease and evaluate ECG sequences very effectively. Hence, the GRU-MLP model is more appropriate for this application since it gives higher accuracy and reliable predictions about heart disease. The suggested methodology attains the maximum level of accuracy at 99.5%. This work propels the field of medical diagnostics by showing the utility of complex neural network topologies in enhancing the predictive power and facilitating the early detection and treatment of heart disease.