Intelligent Ransomware Detection Using Hybrid CNN-Bidirectional GRU with Optimized Training and Low Computational Overhead

The significant increase in ransomware assaults, which peaked over the past decade till 2024, makes it extremely concerning for cyber experts to track early detection methods continuously. This ransomware virus remains one of the most significant threats governments and businesses must confront. Conventional signature-based anti-ransomware solutions and heuristic-based and rule-based methods often struggle to identify ransomware malware, which is ineffective at detecting known threats. Various researchers used machine learning techniques for ransomware detection, leading to a lack of reliability in real-world scenarios and higher computational time costs. To tackle the challenge of ransomware detection, this research work focuses on a deep learning-based hybrid model that combines CNN-LSTM, CNN-GRU, and CNN-Bidirectional GRU. Each layer effectively trains the parameters to detect malware. The CNN-Bidirectional GRU model achieved a maximum accuracy of 99.8% when using the Adam and RMSProp optimizers, with a computational cost of only 0.01 seconds. Using these optimizers, the proposed model reached a greater convergence rate, which protects the files against Ransomware. Additionally, comparisons were made between traditional machine learning and deep learning methods across various metrics, including training accuracy, validation accuracy, training losses and validation losses, to evaluate the overall performance of the proposed methods.