Personalized Health Anomaly Detection in IoT-Integrated Remote Monitoring Systems Using LightGBM and Bayesian Networks

Personalised health anomaly detection is essential in the advancing field of remote patient monitoring via IoT-based devices. Conventional models frequently encounter difficulties with generalised assumptions, resulting in false positives or overlooked detections attributable to individual physiological variability. A hybrid framework integrating LightGBM and Bayesian Networks is presented to attain high-precision anomaly detection with real-time responsiveness. LightGBM effectively identifies feature interactions and nonlinear connections, whereas Bayesian Networks improve probabilistic reasoning in uncertain and dynamic health scenarios. The system was verified using a publicly accessible smart health dataset from Kaggle, attaining an accuracy of 98.7%, precision of 98.4%, recall of 98.2%, F1-score of 98.3%, and an AUC score of 0.996. The suggested model outperforms baseline approaches such as SVM, KNN, and Random Forest in all principal measures. These findings validate its strength and appropriateness for scalable, real-time, and individualised healthcare applications. The suggested method improves patient safety, reduces false alarms, and establishes a new standard in intelligent remote health monitoring systems.