A novel autoencoder based feature independent GA optimised XGBoost classifier for IoMT malware detection

Abstract

The Internet of Medical Things (IoMT) has a network of interconnected medical devices to capture patients' health metrics and store them in a centralized server for analysis by medical experts. The security concerns in IoMT data are therefore very high. The attackers may inject malware into the IoMT data during its transmission. If the health parameters are affected by malware it may mislead the medical experts in making inferences about the patient’s health. IoMT devices are resource-constrained and require faster analysis of data for better medical assistance. Most of the classification techniques used earlier suffer from exhaustive time and resource consumption. Hence, developing an intelligent framework that could reduce the data size and classify the malware quickly is essential. This paper proposes a deep learning framework called Auto-encoder to encode the IoMT data. The encoded features are then given to an XGBoost Classifier whose hyperparameters are optimized using the Genetic Algorithm. XGBoost Classifier detects the presence of malware in the IoMT dataset and Clamp dataset with an accuracy of 98.98% and 98.69% respectively. This lightweight model achieves dimensionality reduction through Autoencoder and effectively detects malware through an optimized XGBoost classifier with faster convergence and limited computational costs. Implementing this framework on the fog layer provides promising results by optimizing resource usage and saving time compared to previous works. Moreover, the proposed method does not need explicit feature selection from the dataset when making predictions.