Integrating LSTM and LOF: A Comprehensive Approach to Anomaly Detection in Healthcare Data

Abstract

The Internet of Medical Things (IoMT) has grown substantially, facilitating extensive time series data accumulation within healthcare environments. Detecting anomalies within IoMT time series data is critical for identifying potential health hazards and ensuring patient safety. This study investigates the efficacy of merging Long Short-Term Memory (LSTM) neural networks with the Local Outlier Factor (LOF) algorithm for anomaly detection in time series data. LSTM networks are adept at grasping extended dependencies in sequential data, while LOF represents a potent unsupervised outlier identification technique. We introduce a unique methodology that capitalizes on the strengths of both LSTM and LOF to heighten anomaly detection accuracy. The proposed technique undergoes assessment via experiments on actual IoMT datasets including WUSTL-EHMS and Thyroid\_Diff, demonstrating consistent performance across diverse healthcare scenarios. A real-time simulation was conducted to assess the feasibility of deploying the framework in practical IoMT environments. Through comprehensive experimentation and analysis, we show its effectiveness. The results of the proposed method reveal a promising ability to precisely pinpoint anomalies, offering a valuable resource to healthcare professionals to quickly identify irregular patient conditions.