LSO-CSL: Light spectrum optimizer-based convolutional stacked long short term memory for attack detection in IoT-based healthcare applications

Abstract

One of the new technological issues of this century is cybersecurity in the health domain. Numerous types of research have been undertaken recently to identify weak points and gaps in healthcare facilities’ and hospitals’ cyber-resilience. Also, different evaluation techniques are used to identify cyber-defense tactics and mitigation methods. During a crisis, however, the health sector is not immune to scientific or professional advancement since it must concentrate all of its resources to treating patients and saving lives. Thus, we proposed Multi Step Convolutional neural network Stacked Long short term memory architecture (MSCSL) for attack detection in Internet of Things detection (IoT) based healthcare applications. Here, using Light Spectrum Optimizer (LSO) the MSCSL’s hyperparameter is optimized. The designed activation function combines the best features of the rectified Liner Unit (ReLU), which is more data-adaptive and has a lower computation complexity. In MSCSL architecture the ReLU is replaced with the Leaky Learnable ReLU (LeLeLU) to improve the data adaptability and minimize the computational complexity. The proposed model is evaluated based on different datasets such as the ECU-Internet of Health Things (IoHT), TON-IoT, and Supervisory Control and Data Acquisition International Electrotechnical Commission (SCADA IEC 60870-5-104) datasets. This dataset includes port scans, brute force, and DoS attacks. When evaluated using different metrics, the proposed method achieves a precision of 98.5%, accuracy of 98.85%, F1-score of 98.74%, recall of 96.3%, Kappa score of 98.9%, and Mathew Correlation coefficient (MCC) values of 0.92, which is higher compared to other existing techniques.