Quantum Mayfly Optimization with Encoder-Decoder Driven LSTM Networks for Malware Detection and Classification Model

Malware refers to malicious software developed to penetrate or damage a computer system without any owner's informed consent. It uses target system susceptibilities, like bugs in legitimate software that can be harmed. For dealing with the new malware, new approaches have been developed to identify and prevent any damage caused. The recent advances in Deep Learning (DL) models are useful for malware detection because they are trained via feature learning instead of task-specific approaches. This paper presents an Optimal Encoder-Decoder Driven LSTM Networks for Malware Detection and Classification (OELSTM-MDC) technique. The presented OELSTM-MDC technique involves the identification and classification of malware. To accomplish this, the OELSTM-MDC model applies pre-processing in the initial stage for data normalization. In addition, Quantum Mayfly Optimization-based Feature Selection (QMFO-FS) approach is derived from choosing an optimal subset of features. Finally, the Butterfly Optimization Algorithm (BOA) is employed for optimal hyperparameter tuning of the ELSTM model. A wide range of empirical analysis is investigated on benchmark datasets to assess the better malware classification performance of the OELSTM-MDC model. It is also compared with the conventional machine learning models such as Random Forest, XGBoost, support vector machine, etc. According to the comparison studies, the OELSTM-MDC model outperformed conventional techniques by detecting the malware class and benign class with accuracy of 97.14% and 98.33% based on the training and testing datasets.