Temporal Convolution Network with Long Short-Term Memory based Predictive Diagnosis for Personalized Healthcare

In modern days, personalized healthcare systems have become crucial in modern medical technology with the growing demand for continuous and real time patient monitoring. Additionally, the integration of Internet of Things (IoT) and Machine Learning (ML) techniques enabled remote monitoring of physiological data, which helps in timely intervention for early diagnosis. However, the existing Convolutional Neural Network (CNN) with Attention Network based health detection system faced several challenges. These limitations include lack of ability to capture temporal dependencies and adaption over time or pa tient history integration. Therefore, the proposed Temporal Convolution Network with Long Short-Term Memory (TCN-LSTM) based personalized healthcare overcome these limitations. Initially, the real time data including Electrocardiogram (ECG) signals, heart rate, body temperature and are collected through various IoT sensors. Next, these data are preprocessed by using various techniques including linear interpolation for handling missing values in vital sign data and z-score normalization for normalizing all the signals. Furthermore, to eliminate the amplitude bias, the collected ECG data is segmented into 10second windows. Lastly, the preprocessed data is fed to the TCN model for extraction of temporal patterns from ECG windows, allowing early prediction of heartbeats and further, LSTM model for forecasting the abnormal conditions such as fever, hypoxia or bradycardia. Lastly, a final health risk classification is generated by concatenating the outputs by using feature concatenation and passing through the fully connected layers.