Derin öğrenmeyi kullanarak zararlı yazılım sınıflandırması

In this study, the performance of different machine learning algorithms was evaluated for detecting malware based on the permissions used by Android applications. The study aimed to compare the performance of four different algorithms: k-nearest neighbors (KNN), a convolutional neural network (CNN), a convolutional neural network with long short-term memory (CNN-LSTM), and a convolutional neural network with the gated recurrent unit (CNN-GRU) by using the Python programming language. The dataset used in the study consisted of binary vectors representing the permissions used by each application and was labeled as either malware or non-malware (benign). The performance of the algorithms was evaluated using standard metrics such as ACC, PREC, and recall. The results showed that the (CNN-GRU) algorithm had the highest accuracy with 95%, followed by the (CNN) algorithm with 91%. The (KNN) algorithm had an accuracy of 87% while the (CNN-LSTM) algorithm had an accuracy of 90%. All four algorithms had similar precision and recall scores, with the (CNN-GRU) algorithm having the highest F1-score of 94%. The loss values for all four algorithms were also calculated and compared. In conclusion, the results indicate that the (CNN-GRU) algorithm performed best in terms of accuracy and F1-score, followed by the (CNN) algorithm. These findings can help inform future research in the field of malware classification using deep learning.