A Learning-based Method for Combining Testing Techniques

This work presents a method to combine testing techniques adaptively during the testing process. The method intends to mitigate the sources of uncertainty of software testing processes, by learning from past experience and, at the same time, adapting the technique selection to the current testing session. The method is based on machine learning strategies. Offline strategies are used to take into account historical information about techniques performance collected in past testing sessions; online strategies are used to adapt dynamically the selection of test cases to data observed as the testing proceeds. Experimental results show that the technique performance can be accurately characterized from features of testing sessions by means of machine learning algorithms, and that integrating this result into the online algorithm allows improving the fault detection effectiveness with respect to single testing techniques as well as to their random combination.