.png){kind=icon}
Return
Optimization of software vulnerability detection model based on machine learning algorithm
.png){kind=icon}
PRE
AI
DOI：10.1007/s00521-026-12447-8.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Conventional methods for detecting vulnerabilities tend to have serious drawbacks leading to high rates of false positives and false negatives alongside with low levels of efficiency of detection and lack of flexibility in dealing with new kinds of vulnerabilities. In order to solve this issue, in this paper the author develops a software defect detection technique based on the methods of the convolutional neural networks and long-short term memory. First of all, Clang and PyLint are used to extract the control flow graph, data flow graph, and variable usage of the source code. Then, random forests and gradient boosting trees are used for vulnerability classification, and detection performance is improved by combining multiple basic classifiers. In addition, this paper combines a convolutional neural network and a long short-term memory network to improve the model’s ability to learn complex code structure and potential vulnerability patterns. To improve model performance, grid search and Bayesian optimization are used to tune hyperparameters. To avoid overfitting, dropout is used, and feature selection based on principal component analysis and L1 regularization is performed to reduce the impact of redundant features. The experimental findings show that the machine-learning model proposed in the paper has improved performance (efficiency, accuracy, and false-negative rate) for identifying vulnerabilities compared to conventional methods. Machine learning applications achieve accuracy > 0.80 in most cases, using no more than 13,000 MB of memory per case analyzed. The model shows much greater flexibility and reliability when analyzing large code bases and different categories of vulnerabilities, thereby improving the overall effectiveness of the detection process and providing an effective means to identify and correct software flaws. This has great theoretical and practical importance.
Keywords:
Software vulnerability detection
Convolutional neural network
Long short-term memory
Machine learning
Feature selection
Journal
IF:
4.5
Papers:
929
Citations:
3.2W
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSaveAn intelligent framework for attack detection in the internet of things using coati optimization
PREAI
ShareSave
ShareSaveShareSaveShareSaveShareSaveShareSave
View more