A Malicious JavaScript Code Detection using Deep Learning Approach on Imbalanced Dataset
Year: 2025
Type: article
Abstract: A web application refers to a software or computer program that makes use of a web browser. In the present day, the categorization of JavaScript-based attacks has significant importance, primarily driven by the substantial increase in the number of Internet users. JavaScript is a crucial component in the field of web development, serving as a major instrument for hackers to initiate attacks. When the presence of malicious JavaScript is identified within Web sites, the rate of detection is seen to rise while the rates of false-negative and false-positive outcomes are observed to decrease. Most of the existing work validates their approaches using balanced datasets which can not be generalized to real-world scenarios. In this paper, we employed GPT-2 as data augmentation technique to address the class imbalance problem. The resultant balanced dataset is then utilized as input for the DOC2Vec algorithm, which facilitates the process of vectorization. The resulting features are then fed into an attention-based Bi-LSTM model. The results demonstrate that the proposed model achieves superior performance in terms of recall and f1-score, with values of 0.74% and 0.50% respectively
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4415441429
