Research On Internet Of Things Malware Classification Algorithm Based On Deep Learning

Internet of Things(Io T)devices built on different processor architectures have increasingly become targets for adversarial attacks.This article proposes an algorithm for the classification of malware in the Internet of Things field to deal with the increasingly severe Internet of Things security threats.Application execution is represented by a continuous sequence of API calls.Based on the improved information gain sub-sequence feature extraction method,the API call sequence of sample data is analyzed and filtered.According to the experimental results,it can effectively reduce the sequence length of the input sample data while retaining important information in the sequence data,which is more effective than chi-square statistics.Subsequently,we proposed an IG-CNN classification model based on a multi-layer convolutional neural network to classify various types of Io T malware,which is suitable for processing time series data.When the convolution window slides down the time series,it can obtain higher-level positions by collecting different sequence features,so as to understand the characteristics of the corresponding sequence position.In order to allow the neural network to increase the convergence speed during the training process,we implemented the random variance reduction gradient optimization algorithm based on Tensor Flow.Comparing the iterative efficiency of different gradient optimization algorithms,the random variance reduction gradient optimization algorithm can significantly accelerate the convergence rate of IG-CNN during training.Comprehensive iterative efficiency and classification performance,we chose an algorithm that can approximate the optimal solution to a small number of iterations to speed up the convergence speed of model training,while ensuring high classification accuracy.The experimental results of the classification of Io T malware families in the real world show that the classification accuracy of this method can reach more than 98%,and it has a certain degree of anti-interference ability.In general,our proposed method has undergone a comprehensive evaluation,showing the practical applicability of the classification of Io T malware families with high accuracy and low computational overhead,and has a great application prospect.