A HYBRID DEEP ENSEMBLE FRAMEWORK FOR INTELLIGENT FAULT DETECTION AND ANOMALY DIAGNOSIS IN MEDICAL IOT SYSTEMS

Abstract

The widespread adoption of Internet of Things (IoT) devices in healthcare has underscored the need for intelligent systems to accurately classify device operational states. With over 60% of hospitals utilizing IoT for patient monitoring and the global medical IoT market projected to reach USD 254.2 billion by 2026, ensuring device reliability is critical. However, studies indicate that approximately 15% of medical IoT devices suffer from undetected malfunctions due to inefficient classification systems. Traditional manual monitoring methods are prone to errors, cannot handle large-scale real-time data, and often miss transient faults that can compromise patient safety and disrupt hospital operations. This study introduces a hybrid classification framework that combines an Artificial Neural Network (ANN) with an Extra Trees Classifier (ETC) to classify device states as either Normal or Anomaly. Data is sourced from hospital telemetry logs and the open-source Medical IoT Device Dataset (MIDD), undergoing preprocessing steps including null value removal, min-max normalization, and time-series segmentation. A baseline Gaussian Naïve Bayes Classifier (NBC) demonstrated moderate accuracy but failed to capture nonlinear relationships. In contrast, the ANN enables deep temporal feature extraction, and ETC ensures robust and efficient classification. The ANN with ETC model significantly outperforms traditional approaches in accuracy and anomaly detection, offering a reliable solution for real-time medical IoT monitoring