Verification of the effectiveness of fuzzy rule-based fault prediction: A replication study

The prediction success of faulty modules in a software helps practitioners to plan the budget of software maintenance that leads developers to improve the reliability of software systems. Despite various learning algorithms and statistical methods, fault prediction needs novel methods for enhancing the success of the prediction. Fault prediction can be performed using fuzzy rules that are new for this field. In this work, fuzzy rule-based fault prediction approach, which was developed by Singh et al. [11], is replicated to validate the success of fuzzy rule-based fault prediction in open-source data sets. The steps of the experiment and the steps of Singh et al's work, which are applied for replication, both are same. Classification is performed after generating clusters that are constituted using fuzzy rules in normalized data sets. According to the prediction results obtained by applying 10*10 cross-validation, fuzzy rule based fault prediction produces less errors in open-source data sets when it is compared with industrial data sets. In addition to this, the results validate the findings of Singh et al.'s work in terms of some performance parameters of the fault prediction.