Software Faulty Execution Diagnosis Using Clustering

The aim of this study is to suggest a method for high accurate and automatic fault detection in the software testing phase without direct inspection to the source code. Current approaches to this problem try to discover and introduce suspicious points in the code by comparing the several execution times of the program and inspecting some specific points in the program, defined for this propose called predicate . However, the problem is that analysis of run-time data of predicate point ends up to inefficiency of current methods in detecting various faults without considering the relation between them. In order to take these problems into account, the present study introduces a new solution on the basis of run-time data analysis of execution path behavior. Therefore, possible relations among predicates are taken into consideration. Also, a clustering method is used to detect the faulty execution which is similar to a passed execution more than any other one so that a possible fault can be located. Compared to previous methods, the proposed method shows success in the detection of faults by means of a set of real dataset and the application thereof for a number of real programs.