New Trials on Test Data Generation: Analysis of Test Data Space and Design of Improved Algorithm

Test (input) data generation is important for the algorithms/software/hardware testing. The previous researches on test data generation motivate us to find some meaningful information of generated test data. In this paper, we differentiate the test (input) data space (i.e., problem instance space) from the output data space (i.e., solution space), by examining the test data generated in terms of optimality (one of the performance measures). We investigate the problem instance space of the 0/1 knapsack problem, by calculating some kind of cost-distance correlation; the correlation was quite different from those in well-known combinatorial optimization problems. Also, we improved a greedy algorithm of the 0/1 knapsack problem by generated test data. The improved algorithm showed superiority to the original one under 10,000 random instances. This paper presents some promising values of the researches on the test data space and the test data generation for improving the tested module.