Software Susceptibility Detection Using Deep Learning Techniques

Software susceptibilities are flaws in the source code that could be used to do havoc or collect information. A variety of automated machine learning methods have been suggested in prior research to find them. The majority of these methods operate at the release level, which means they try to foresee which files might be vulnerable in a future version. Manually detecting insecure code is both challenging and costly in both time and money. Therefore, to reduce costs, developers must use automatic vulnerability prediction (AVP) techniques. Recent AVP research has started to employ deep learning (DL) methods. All of the suggested methods are built on feature extraction techniques that were influenced by earlier DL uses such as autonomous language processing. When developing AVP models based on traditional machine learning, code metrics were frequently employed as features. The following piece provides a method centered on deep neural networks over leveraging code metrics to detect vulnerabilities in code, bridging the difference between machine learning parameters and deep learning. Results obtained indicate that while code metrics are excellent, they weren't the best to use as DL-based AVP characteristics.