Data Analysis and Pattern Recognition for Software Anomalies

In this paper, we report on the preliminary results of a study that is underway to examine the software problem failure report data associated with a diverse set of NASA projects for insights into the processes, lessons learned for improving them, and predictive models to help determine the expected metrics for a given project based on key attributes. We report on the patterns that are found to date by analyzing the software problem failure reports data for several NASA mission types. These patterns help determine the underlying factors and root causes of problems and preferred methods for mitigating them. The current study is a broad review of the data to support the clear classification and formulation of the techniques to be used for more in-depth analysis. In the future, we plan on building models to predict anomaly rates based on mission characteristics. Preliminary observations and recommendations are suggested for improving software assurance activities and upgrading the forms used for recording the errors so that they are easier to use for the experts recording the errors and clearer for the experts analyzing them