Multiscale attention-based deep learning approach for ransomware threats detection and prevention

In this research work, a new deep learning-based ransomware threat detection and prevention model is developed to protect the files in digital computing devices. Initially, essential data is collected from benchmark sources, and it is then subjected to the feature extraction phase, where Statistical features, Principal Component Analysis (PCA)-based features, and deep features are extracted. In addition, optimal features are selected using Enhanced Arbitrary Variable-based Nuclear Reaction Optimization (EAV-NRO). Finally, the extracted ensemble features are fed into the designed Multiscale Attention-based Residual Long Short-Term Memory (MA-ResLSTM) model to perform ransomware threat detection. The efficacy of the developed mechanism is verified among different state-of-the-art approaches. In dataset 1, the developed MA-ResLSTM approach attained 93.5% accuracy and 93.46% sensitivity, which is better than MLP, GAN, RNN and dual-layer RF. Thus, the outcome proved that the accurate detection of developed ransomware attack frameworks ensures that real threats are identified and mitigated properly.