Improving the Effectiveness of Fault Localization Techniques Using Fuzzy Expert System

• Main Authors: Wen-Yuan Chen, 陳玟瑗
• Other Authors: Lin, Chu-Ti
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/wv6f4a

碩士 === 國立嘉義大學 === 資訊工程學系研究所 === 107 === Software debugging includes the following steps: locating the fault, determining the root cause, designing and implementing the code fix. Locating faults genereally takes a lot of time. In the literature, many spectrum-based fault localization (SFL) techniques have been proposed in order to improve debugging efficiency. Those SFL techniques were designed according to different underlying assumptions and then evaluate the probability that each statement contains faults (called suspiciousness). Finally, SFL will schedule the statements in an order based on the susciciousness values so that the statements with higher suspiciousness values will be examined by software developers early during the debugging process. However, since different SFL techniques are proposed based on different underlying assumptions, their capability to locate different faults are different. If developers adopt a single SFL technique, some faulty statements may be assigned low suspiciousness values in specific conditions, thus resulting in worse debugging efficiency. More specifically, no SFL technique’s underlying assumptions conforms to the causes of all faults. Due to the aforementioned reasons, we propose a framework for improving SFL effectiveness by using the fuzzy expert system to integrate different SFL techniques. In the proposed framework, the outputs of several SFL techniques will first be put into the fuzzification and will be transferred to fuzzy input sets by means of membership functions. After undergoing the Mamdani fuzzy inference based on some fuzzy rules, the fuzzy sets will be transferred to fuzzy output sets. Finally, the fuzzy sets will be transferred to cript outputs (called weighted suspiciousness values) by means of the center of gravity (COG) method. The statements will then be ranked according to their weighted suspiciousness values. Our empirical results indicatethat the proposed framework can achieve better SFL efficiency thanh the existing SFL techniques that are compared in this study.