Combi-FL: Neural network and SBFL based fault localization using mutation analysis

Abstract

In this article, we present a hybrid approach for fault localization (FL). We combine three different domains of software FL techniques to realize an effective fault localizer. Spectrum based fault localization techniques and neural network (NN) based techniques are utilized to determine the similarity of different mutants with the faulty program. Ten prominent FL techniques (six techniques from SBFL family and four methods from NN family) are considered in our proposed approach: Combi-FL. The ranking sequences generated by different FL techniques are combined using learning-to-rank algorithm. In this work, we focus on localization of single-fault programs. We have evaluated our proposed Combi-FL technique over seven program suites comprising twenty-one programs and compared its effectiveness with eight popular FL techniques. Our experimental results show that on an average Combi-FL is 28.42% more effective than existing FL techniques such as DStar, Ochiai, Barinel, Tarantula, CNN-FL, DNN, RBFNN, and BPNN.