Malware Detection Using Convolutional Neural Network and Perceptron Neural Network Optimized with Firefly Algorithm

Abstract

Abstract—In this paper, in order to detect 25 classes of malware, with the aim of increasing the detection accuracy, we used the pre-trained convolutional neural network of Alex Net and combined it with the perceptron neural network optimized with the Worm Shabbat algorithm.  In fact, Alex Net’s convolutional neural network automatically extracted 1000 feature vectors for each input image using the convolutional layer in its architecture.  In the next step, we used the transfer learning method to classify the extracted features.  In this thesis, we transferred the learning done by the Alex net convolutional neural network to a multi-layer perceptron neural network that was optimized using the firefly meta-heuristic algorithm for classification.  In this work, we optimized the optimal weight and bias of the neural network by meta-heuristic algorithm.  Finally, we were able to achieve 99.8% accuracy, which showed that the proposed method was superior in terms of accuracy compared to the compared methods.