Performance Analysis of LSTM, SVM, CNN, and CNN-LSTM Algorithms for Malware Detection in IoT Dataset

Machine learning is an effective technique to tackle both the detection and classification tasks of malware. This is realized through learning algorithms that use various distinguishing features that characterize malware. Today's malware uses extremely sophisticated techniques, which means that various techniques to combat it are intensively developed. When malware is invisible, it can compromise many different data of a large number of users. Therefore, it is necessary to first analyze the types of malicious software and then propose appropriate countermeasures. In this regard, this work aims to analyze the performance of some well-known machine-learning techniques based on neural networks and support vector machines, originally developed as a method for the efficient training of neural networks. For the goal SVM, LSTM, CNN, and CNN-LSTM algorithms are analyzed concerning their effectiveness in the classification of malware in IoT datasets. For all the algorithms studied, their confusion matrices are presented along with receiver operating characteristic curves. The best results were obtained using the hybrid CNN-LSTM approach. Its results showed an accuracy of 97% and balanced performance across all metrics.