A Two-Stage Anomaly Detection Framework for Improved Healthcare Using Support Vector Machines and Regression Models

This paper presents an effective two-stage anomaly detection technique for wireless body area networks (WBANs)-based personalized healthcare monitoring. Our method uses SMO regression (Sequential Minimal Optimization) for more in-depth contextual analysis and Support Vector Machines (SVM) for initial anomaly categorization. Domain-specific thresholds and sophisticated feature engineering are utilized to accurately identify point anomalies brought on by sensor noise or malfunctions and contextual anomalies resulting from interdependent physiological feature deviations. The robustness of the model is improved by features including heart rate discrepancies, rolling mean, and rolling standard deviation. With 99.61% and 99.97% detection accuracy for point and contextual anomalies, respectively, the model attains an overall accuracy of 99.59%. These findings highlight the promise of scalable, customized anomaly detection systems for WBAN-driven, real-time healthcare, which could lead to prompt treatments and better patient outcomes.