MalGA-LSTM: a malicious code detection model based on genetic algorithm optimising LSTM trainable parameters

With the development of internet technology, the number of malicious software is also growing rapidly, causing great potential for cybersecurity issues. When using neural network to identify and detect malicious code, the traditional gradient descent method is easy to fall into local optimum and sensitive to the initial weight of the network. In order to solve these problems, a method using genetic algorithm (GA) to optimise LSTM trainable parameters for malicious code detection is proposed in this study. First, the API sequence called by malicious code was transformed into word2vec word vector, then genetic algorithm was used to optimise the trainable parameters in the network. The experimental results showed that the accuracy of the LSTM model optimised by genetic algorithm in the training set was more than 15% higher than that of the traditional gradient descent method, reaching 94.53%, and the accuracy in the testing set was more than 10% higher than that of the traditional gradient descent method, reaching more than 86%. © 2023 Inderscience Enterprises Ltd.. All rights reserved.