Fault Localization Based On Graph Mining

Nowadays,the computer software has involved in all fields of human society.In order to ensure the quality of software,people pay more attention to software testing.Program debugging is a difficult task in software testing.Fault localization is a laborious step in program debugging.Improving the efficiency of software fault localization can effectively reduce debugging costs.Therefore,the study of effective fault localization methods is of great significance to improve the efficiency of program debugging,reduce the cost of software testing and ensure the quality of software.Many researchers at home and abroad have proposed different kinds of approaches for fault localization,but these methods ignore some program execution statistics,and the final results lack fault-related context information.Based on the analysis of existing methods and techniques,this paper proposes a method for fault localization based on graph mining and support vector machine,and a method for fault localization based on chemical reaction optimization,finally implements a tool for fault localization based on chemical reaction optimization.To overcome the limitation that existing methods ignore some program execution statistics in fault localization,this thesis introduces a method for fault localization based on graph mining and support vector machine.The proposed method first collects the trace of program execution and models it into the software behavior graph;then the reduced behavior graph is used to construct the weighted software behavior graph,and the graph mining algorithm is used to mine the closed subgraph and record the frequent edges;finally,the software behavior graph is transformed into feature vector,and support vector machine is used to classify all graphs,and methods which improve the classification accuracy are added into the set of suspicious methods for fault localization.The experimental results show that this method can improve classification accuracy and promote understanding of failure generation.To solve the problem that the failure context information is missing for fault localization results provided by the existing methods,this thesis introduces a method for fault localization based on chemical reaction optimization.The proposed method first collects the program execution traces to build the software behavior graphs and reduces them;then,the context is encoded according to the behavior graphs,and the molecular structure is defined to search the optimal molecular population through the chemical reaction optimization;finally,the molecules in the population are transformed into the context for fault localization.The experimental results indicate that this method can provide context information of failure generation and promote understanding of execution failure.On this basis,we design and implement a tool FLUCRO for fault localization based on chemical reaction optimization using Java,which provides a visual interface for parameter configuration.In summary,aiming at some problems existing in current methods of fault localization,this paper proposes two methods for fault localization,and implements a tool to help developers locate faults quickly.