Research On Multi-objective Restricted Boltzmann Machine Model For Malicious Code Detection

With the rapid development of big data and the Internet,the existence of malicious code also poses a serious threat to network security.To this end,a large number of scholars have researched malicious code detection from different levels,including physical technology,semantic analysis,deep learning methods,and so on.However,in the face of huge malicious code datasets,the use of inappropriate data processing methods not only reduces the efficiency of malicious code detection,but may lead to a trend of anti-growth of malicious code.Therefore,in response to the above problems,this article aims to discuss improving the detection effect of malicious code,and proposes two different deep learning models,namely the multi-objective restricted Boltzmann machine model and the multi-objective convolution restricted Boltzmann machine model based on the spatial pyramid pooling strategy.Moreover,this article proposes a fast non-dominated sorting genetic algorithm based on the constrained dividing crossover strategy from the perspective of the algorithm to optimize the imbalance malicious code datasets.The main works of this article are as follows:In order to reduce the impact of imbalanced datasets on malicious code detection,this paper uses the intelligent optimization algorithm NSGA-II to optimize the imbalanced malicious datasets,and designs a novel multi-objective RBM model with optimized parameters.The main idea is to obtain the optimal training solution through NSGA-II algorithm,and input to the multi-objective RBM model for training and testing to achieve the classification of the malicious code image dataset,and then use two objective functions to evaluate the optimization effect of the imbalanced data and the performance of data classification.This paper conducts experiments by comparing different deep learning models,dataset processing methods,and image resolution.The experimental results show that the learning ability of imbalanced datasets is enhanced and better data classification results are obtained.At the same time,the robustness of the multi-objective RBM model is verified,which pave the way for malicious code detection.Taking into account the limitations of the RBM model when processing a large number of image datasets and CNN has the advantages of faster training speed and better classification effect,therefore,in order to further improve the effect of data classification and better realize malicious code detection,we design a multi-objective CRBM model based on SPP strategy.The model mainly introduces the SPP strategy in the last pooling layer and uses the parameter-optimized RBM model in the fully connected layer as a generative model.Then three objective functions are combined to evaluate the optimization effect of imbalanced datasets and the effect of model training to achieve data classification.The processing method of imbalanced dataset is also optimized by NSGA-II algorithm.Compared with the multi-objective RBM model,the multi-objective CRBM model shows further advantages in the processing of imbalanced datasets and data classification.Extensive experimental results show that the proposed model has better performance.Since the training of the imbalanced dataset restricts the detection effect of malicious code to a certain extent,thus,according to the characteristics of the imbalanced dataset,this article improves the NSGA-II algorithm,and an NSGA-II algorithm based constrained dividing crossover strategy is proposed to optimize the imbalanced dataset;and then two objective functions are designed to evaluate the optimization effect of the imbalanced dataset;finally,extensive comparison experiments are conducted on the standard test set.The simulation results indicate the effectiveness and feasibility of the CDCS-NSGA-II algorithm in optimizing imbalanced datasets,which also provides new ideas for handling imbalanced datasets in the process of malicious code detection.