Novel scheme to locate software fault by aggregate analysis of program behaviors

It is an interesting and noteworthy issue to locate rapidly and effectively the fault according to the input and the program behavior in case of software failure. A drill-down framework for fault localization was proposed, based on which the fault localization problem was expressed in a step-by-step procedure with multi-hierarchy. Proposed and analyzed is a novel scheme of fault-locating method on input level using aggregate analysis of the program behavior. Statistical model of the fault-locating approach on input level was established and the corresponding solution were presented and discussed. To validate the proposal, an experiment was conducted, involving 5 software, 4 of which are injected with faults deliberately and the rest one is actual software in use. The results show that an average precision of 0.803 and an average recall of 0.697 are obtained. Finally, the application of aggregate analysis of the program behavior into the fault-locating method on module level was discussed.