Adequate Test Data Generation using Evolutionary Algorithms

Abstract

Software Testing is a approach where different errors and bugs in the software are identified. To test a software we need the test data. In this thesis, we have developed the approach to generate test data automatically from some initial random test data using Evolutionary Algorithms (EA) and test the software to detect the presence of errors, if any. We have taken two measures, they are path coverage and adequacy criterion to test the validation of our approach. In our first approach, we have used simple Genetic Algorithm (GA) to find the test data. We then used an memtic algorithm to curb the difficulties faced by using GA. We are using the instrumented program to find the paths. We then represent the program into a Control Flow Graph (CFG). We have used genetic algorithm to find the more optimal test data that covers all the feasible test paths from some initial random test data automatically. Path coverage based testing approach generates reliable test cases. A test case set is reliable if it's execution ensures that the program is correct on all its inputs. But, Adequacy requires that the test case set detect faults rather than show correctness. Hence, for adequacy based testing we uses the concept of mutation analysis. Here, we have taken the mutation score as our fitness function in the approach. We find out the mutation score from using mutation testing based tool called "MuJava". And then generate test data accordingly. We applied a more complex hybrid approach to generate test data. This algorithm is a hybrid version of genetic algorithm. It produces better results than the results generated by using GA. Also it curbs various problems faced by GA