An optimal insider threat detection model based on improved deep belief network with feature reduction scheme for e-healthcare system

An electronic health records (EHR) dataset contains routine actions performed when accessing a patient's record, Patient user information corresponds to the record's identifier, whereas device data describes how the record was accessed. Leakage of valuable information is a critical challenge. As the Internet-of-Things (IoT) evolves, new security challenges arise in existing security architectures. An organisation's insider threat management is at risk because attack surfaces have expanded dramatically. To solve the above challenges, an optimised deep belief network (DBN) is proposed to detect insider threats in EHR. Significant features are generated using correlation coefficients, random forest mean reduced accuracy, and gain ratio to improve the performance of the internal threat detection model. An appropriate mechanism (and function) is then used to combine the features to obtain an optimal set of features. Adaptive rat optimisation algorithm (AROA) optimises DBN weight parameters to enhance performance. F-measure, accuracy, and G-mean are calculated to measure Performance.