Software fault prediction using cross-validation

Abstract

Software faults are dangerous. Software systems are often essential to a business operation or organization, and failures in such systems cause disruption of some goal-directed activity (mission critical). Faults in safety-critical systems may result in death, loss of property, or environmental harm. Run-time faults are the most damaging as they are not always detectable during the testing process. Detecting faults before they occur gives the designers a brief inner view of the possible failure and their frequency of appearance. This helps in focused testing and saves time during the software development. Prediction models have the ability to differentiate between various patterns. This article showcases the effectiveness of cross-validation in design and development of a neural network for software fault detection.