CLOUD-BASED HEALTHCARE MONITORING SYSTEM FOR ABNORMAL HEALTH PATTERN DETECTION USING CONVLSTM

Abstract

The healthcare domain is increasingly adopting technology to enhance patient monitoring, diagnosis and treatment through data collection and advanced predictive analytics. Traditional systems struggle with delays in diagnosis and the inability to adapt to continuous changes in patient health, leading to suboptimal care. To address these problems, this work proposes to develop a cloud-based healthcare monitoring system that uses deep learning to detect abnormal health patterns, enabling early detection and timely intervention for potential health risks. The system begins with gathering healthcare data from medical records, wearable devices and IoT sensors. After that, data preprocessing follows, where missing values are handled using K-Nearest Neighbors imputation and outliers are detected using the Interquartile Range method for data consistency. Feature extraction is then performed using Discrete Wavelet Transform to decompose time-series data into frequency components, capturing both time and frequency features. The extracted features are fed into a Convolutional Neural Network with Long Short-term Memory (ConvLSTM) model, which combines convolutional layers to capture spatial patterns and LSTM layers for capturing temporal dependencies to classify the data as normal or abnormal. Finally, the trained model is deployed in the cloud, ensuring scalability, accessibility and continuous model updates through periodic retraining with new data. Results demonstrate that the system achieved an accuracy of 99.22%, precision of 99.34%, sensitivity of 98.58%, specificity of 98.23% and an F-Measure of 97.74%. Additionally, latency of 35 ms at 200 ms response time, highlighting the model’s performance. The proposed approach offers an efficient, scalable solution for continuous patient health monitoring, providing accurate predictions, reducing the time to diagnosis and ensuring timely interventions for improved patient care.