Anomaly detection and prediction of smart elderly care health data by combining multi-task learning and support vector machine

Smart elderly care systems often struggle with accurate anomaly detection and health trend prediction due to the multi-source heterogeneity and high noise characteristics of health data. Existing methods face limitations in handling structural differences across data sources and capturing task correlations. To address these challenges, this study proposes a novel framework that integrates multi-task learning (MTL) with support vector machines (SVMs). Health data are collected from smart wearable devices, electronic medical records, and environmental monitoring sensors. Following data cleaning and normalization, an MTL framework is constructed to extract shared representations between anomaly detection and health trend prediction tasks. An SVM model with a radial basis function (RBF) kernel is employed for robust anomaly detection in high-dimensional data. Additionally, a hierarchical prediction mechanism is developed to dynamically forecast health trends using shared features and classification boundaries. Experimental evaluations on a real-world smart elderly care dataset demonstrate that the proposed method achieves an anomaly detection accuracy of 97% and a mean squared error (MSE) of 0.12 in health trend prediction. These results confirm the effectiveness of the approach in enhancing the analysis of complex health data, offering a promising solution for intelligent data processing in elderly care applications.