A Data Mining Approach to Discover Temporal Relationship among Performance Metrics from Automated Testing

E.M.H. Lee and K.C.C. Chan (PRC)


Automated software engineering, performance testing, performance metrics, temporal data mining, association rule mining, and multivariate time series analysis


Companies seek to ensure software speed and scalability by testing the performance of their systems, generating large volumes of performance data that is presented graphically and analyzed by testers. This is a time consuming process. This paper proposes a way to process these large amounts of data using a machine learning technique from association rule mining. It suggests an improvement to the current automated test tool model, a data-driven model which can be used to automatically find useful knowledge in target data. Domain experts, in this case software testers, can use this knowledge as a reference in data analysis. We experiment this technique with the load-testing results captured in the evaluation of an open source phone directory application by a company. The test results are pre-processed and input into a data mining engine. The analytical method is fast, accurate and discovers interesting rules. We discuss how these results might be used in a practical setting.

Important Links:

Go Back