LSTM-based deep learning for spatial-temporal software testing

Continuous integration (CI) software development practice has become more and more popular. Regression testing occurs very frequently in CI. Test case suites constantly change since new test cases are inserted and obsolete test case are removed in each cycle. The software developer hunts for quick-feedback of faults because of time constraint. An embedded software usually includes the spatial-temporal data in CI. The efficiency of regression testing for the embedded software is related to the space-time. To achieve ideal regression testing goals for the embedded software in CI, this paper proposes a novel test case prioritization approach using LSTM-Based (Long short-term memory) deep learning. LSTM is a time series prediction model. It can predict the probability of each test case detection fault in the next cycle according to the testing history information of all the previous CI cycles. The priority of test case can be obtained dynamically under the guidance of the probability. The experiments are conducted on two industrial data sets. The results verify that compared with some exiting test case prioritization approaches, our approach has better performance for embedded software as follows: (1) improve the prioritization effectiveness, (2) increase the fault detection rate in CI environment, and (3) decrease the testing execution time through automatic reduction the obsolete test cases.