BIOCAD: Bio-Inspired Optimization for Classification and Anomaly Detection in Digital Healthcare Systems

Abstract

The modern smart digital healthcare system (SDHS) is leaning towards automation of patient disease monitoring and treatment with the advent of wireless body sensor networks (WBSN) and the internet of medical things (IoMT). However, the open communication network for sensitive medical data transfer is giving rise to vulnerabilities and security concerns. To prevent adversarial manipulation of sensor measurements, SDHS IoMT controllers leverage anomaly detection systems on top of the disease classification systems. Machine learning (ML) is one of the most effective techniques for providing experience-based automated decision-making models. These models generalize well to produce the expected output for the unseen inputs from the learned patterns. Therefore, ML-based models are currently being adopted to automate the anomaly detection and disease classification tasks of SDHS. In this work, we consider a SDHS that uses supervised ML models for patient status/disease classification and unsupervised ML models for anomaly detection. However, the performance of the ML models largely depends on hyper-parameter tuning. Finding the optimal hyper-parameter is a challenging task, and it becomes more difficult and time-consuming in high-dimensional feature space. In this work, we propose BIOCAD, a comprehensive bio-inspired optimization framework for SDHS data classification and anomaly detection. The framework leverages a novel fitness function for unsu-pervised anomaly detection ML models. We experiment with state-of-the-art datasets - the Pima Indians diabetes dataset, the Parkinson dataset, and the University of Queensland vital signs (UQVS) dataset for validating our proposed strategy.