Anomaly Detection in Wireless Body Area Network using Mahalanobis Distance and Sequential Minimal Optimization Regression

Wireless Body Area Network (WBAN) is a health service that consists of small sensors attached to the patient's body to improve health services. Apart from the importance and necessity of the WBAN, there are some problems in WBAN such as inaccurate measurements, hardware failures, discharged sensors, and sensors running out of energy which can cause false alarms and reduce confidence in using the WBAN system for remote health services. This research aims to create a system or modelling to solve anomalous data problems in WBAN health data. The system will process data to produce an analysis of information in an anomalous condition by applying the statistical method using Mahalanobis Distance and the prediction method using SMOreg. The process of collecting data and pre-processing is carried out before completing the detection of anomalous features. The data normalization process is carried out to make the process faster. This approach was tested on a real health dataset on the WBAN sensor. The results of the implementation show that the method produced a high-performance detection ratio (DR), which is above 80%. In addition, the experiment resulted that the size of the sliding window affected the anomaly detection results.