Identification of Vulnerabilities in the Visualization of Statistical Metrics for Software Using Neural Networks

When solving computer security issues, visualization is a powerful tool that can be used as an auxiliary aid. However, there are some challenges and difficulties associated with its use. For instance, there may be errors and discrepancies in the interpretation of charts by different individuals. To address these issues, applying neural networks to visual analysis could help automate the process and reduce human error. This paper explores the use of neural networks for identifying vulnerabilities in software by analyzing visual representations of software statistics. It compares the effectiveness of anomaly detection on various types of graphs and conducts experiments to evaluate the accuracy of identifying vulnerabilities using neural network models. The goal is to visually detect a vulnerability based on a single value rather than relying on a group of values. The experiments show that a convolutional neural network is able to detect an anomaly in a graph, even at a single point on a graph, instead of an entire cluster of values.