An Analysis of a Signature-based Approach for an Intrusion Detection System in a Wireless Body Area Network (WBAN) using Data Mining Techniques

Abstract

Wireless Body Area Networks (WBANs) use biosensors worn on, or in the human body, which collect and monitor a patient’s medical condition. WBANs have become increasingly more beneficial in the medical field by lowering healthcare cost and providing more useful information that medical professionals can use for a more accurate, and faster diagnosis. Due to the fact that the data collected from a WBAN is transmitted over a wireless network, there are several security concerns involved. This research looks at the various attacks, and concerns involved with WBANs. A real physiological dataset, consisting of ECG signals obtained from a 25-year-old male, was used in this research to test accuracy of various decision tree classifiers. The Weka software was used to analysis the accuracy and detection rate results of this dataset in its original form, versus a reduced dataset consisting of less, more important attributes. The results concluded that the use of decision tree classifiers using data mining, is an efficient way to test the increased accuracy on a real dataset obtained from a WBAN once it has been altered. The original dataset produced results where the ROC curve ranged from 0.313 (31%) to 0.68 (68%), meaning their accuracy is not very high and the detection rate is low. Once an attribute selection feature was used on the dataset, the newly reduced set showed ROC curves ranging from 0.68 (68%) to 0.969 (97%) amongst the three classes. As a result, decision tree models were much more accurate with a higher detection rate when used on a real dataset that was reduced to function better as a detector for a WBAN.