Anomalous Data Detection in WBAN Measurements

Wireless Sensor Networks (WSN) are vulnerable to numerous sensor error and inaccurate measurements. This vulnerability reduces the efficiency of many WSN application, such as healthcare in Wireless Body Area Network (WBAN). For example, faulty measurement from sensor gives a false alarm to healthcare personnel and lead to wrong patient's handling. Therefore, a system to differentiate between real medical condition and a false alarm will improve remote patient monitoring systems and other healthcare service using WBAN. In this paper, a novel approach is proposed to do anomaly detection using prediction method. The objective of this paper is to make a system which can differentiate between real medical conditions and false alarms. This system forecast a sensor value from historic values and compares it with actual data from real measurement. The difference is compared to a threshold value, which is dynamically adjusted. Then using majority voting algorithm to determine whether the data is an anomaly or not. The proposed approach has been applied to real datasets and compares the prediction methods and the size of the sliding window. Experimental results show the effectiveness of the system, indicated by high Detection Rate and low False Positive Rate.