AI-POWERED ANOMALY DETECTION IN IOT-BASED HEALTHCARE MONITORING: A COMPARATIVE ANALYSIS OF MACHINE LEARNING APPROACHES

Abstrak

The Internet of Things (IoT) enables continuous remote monitoring of patient vital signs, but the effectiveness of such systems depends on their ability to detect anomalous health readings in real time. This paper presents a comparative evaluation of machine learning (ML) methods for anomaly detection in IoT healthcare monitoring, benchmarked against a conventional threshold-based approach. We design a five-layer IoT architecture using MQTT protocol and implement four ML algorithms—Random Forest, Support Vector Machine (SVM), Gradient Boosting, and Isolation Forest—alongside a threshold-with-moving-average baseline. Using 1,000 simulated vital sign readings (heart rate, temperature, SpO2) with 67 injected anomalies across five pathological categories, we evaluate all methods on accuracy, precision, recall, and F1-score. Results demonstrate that supervised methods achieve 100% accuracy and F1-score, outperforming the threshold baseline (F1: 95.71%) and unsupervised Isolation Forest (F1: 93.02%). Feature importance analysis reveals that engineered deviation features contribute most to classification, with temperature deviation achieving the highest importance score of 0.2419. The system achieves sub-millisecond latency (0.1037 ms mean) and throughput of 9,604 msg/s.