Test Cost Reduction for 5G and Beyond using Machine Learning

Software testing is essential, but expensive, especially for significant issues, feature-rich systems such as telecommunication systems evolving toward 50 and beyond. There is a need in this domain for effective testing techniques to ensure that a minimal number of test cases assess the most important combinations of system functions with respect to domain-specific criteria. Our approach aims to address this challenge by first automatically mapping existing test cases to the combinations of system capabilities they exercise and visualizing the mappings using decision tree learners. Then, the approach uses a combination of the engineers' feedback (domain-specific criteria), mapping data, and test execution logs to propose new test cases covering newly-added capabilities or better exercising/verifying existing ones while ensuring the efficacy at fault detection, code coverage, equipment cost, test execution time, redundancy avoidance, among other things.