Software Vulnerability Assessment and Classification Using Recurrent Neural Network and LSTM

Abstract

Software defect detection is a valuable tool for enhancing the quality of technology and testing management. It allows for the quick identification of flaws in simulation models before the real testing phase begins. These prediction results assist technology designers in efficiently allocating their limited resources to areas that are more susceptible to shortcomings. This research presents a novel method for software bug prediction using deep learning techniques. A Recurrent Neural Network is used to classify source code, using various soft computing approaches. Data balancing for normalisation has included the use of many pre-processing and data filtering procedures. The generation of the Vector Space Model (VSM) has included the use of TF-IDF and related feature extraction approaches. The classification was performed using Recurrent Neural Networks (RNN) based on the training module for both the training and validation datasets. The proposed deep learning framework comprises many optimisation strategies, each with its own distinct advantages and constraints. We have assessed all methodologies and chosen the most superior one. To conduct the observations and test the suggested technique, a range of real-time and synthetic accessible datasets are assessed. The evaluation findings demonstrate that the proposed framework version surpasses both simple models of outstanding quality and complex deep learning classification models.