Forecasting the Number of Bugs and Vulnerabilities in Software Components using Neural Network Models

Abstract The frequency of cyber attacks has been rising rapidly lately, which
is a major concern. Because each attack exploits one or more vulnerabilities in the
software components that make up the targeted system, the number of vulnerabilities
is an indication of the level of security and trust that these components provide. In
addition to vulnerabilities, the security of a component can also be affected by
software bugs, as they can turn into weaknesses, which if exploited can become
vulnerabilities. This paper presents a comparison of several types of neural networks
for forecasting the number of software bugs and vulnerabilities that will be discovered
for a software component in certain timeframe, in terms of accuracy, trainability and
stability to configuration parameters.