Analyzing the history of software systems to predict class changes

Determining the critical parts of a system is key to effectively conduct preventive software maintenance. To accomplish this task, information from the system history (i.e., past versions) can be helpful for identifying those software elements that are more likely to receive modifications in the near future. However, interpreting the large amount of data usually present in the history can be difficult. In this work, we propose an approach adapted from financial markets for analyzing the history of an object-oriented system and predicting the classes that might change. We have evaluated our approach by comparing it with existing approaches. The results, although preliminary, show that our approach makes more accurate predictions for class changes.