An Artificial Intelligence-Driven Deep Representation Learning Model for Securing Privacy-Preserving Applications in Human–Computer Interface Systems

The rapid progression of Internet connections has led to a significant increase in cyber-attack events, often resulting in severe and disastrous consequences. Malware is one of the primary weapons used to achieve malicious objectives in cyberspace, either through the exploitation of newly discovered vulnerabilities or through the misuse of features introduced by new technologies. The development of more effective and robust malware defense mechanisms is considered a critical necessity in cybersecurity. Human–Computer Interface (HCI) systems have become increasingly important, and as communication, computing, and display technologies continue to advance, conventional HCI methods may become a bottleneck in effectively handling the growing data flow. Federated Learning (FL) is a Machine Learning (ML) paradigm that aims to train models through decentralized devices, whereas the local data are kept private. In this manuscript, a Secure and Efficient Federated Learning using Optimization Algorithms and Deep Learning for Cybersecurity Applications (SEFL-OADLCA) model is proposed for HCI systems. The aim is to present an effective FL framework to address cybersecurity issues. First, the min–max scaler method is applied for data preprocessing. For the Dimensionality Reduction (DR) process, the Mountain Gazelle Optimizer (MGO) technique is employed. Furthermore, a hybrid Temporal Convolutional Network and Long Short-Term Memory (TCN+LSTM) technique is utilized for the attack classification process. Finally, the hyperparameter selection process is performed using the Remora Optimization Algorithm (ROA) to optimize the classification results. The comparison study of the SEFL-OADLCA method portrayed a superior accuracy of 99.46% compared with existing approaches on the NSL-KDD dataset.