The usefulness of software metric thresholds for detection of bad smells and fault prediction

Context Software metrics may be an effective tool to assess the quality of software, but to guide their use it is important to define their thresholds. Bad smells and fault also impact the quality of software. Extracting metrics from software systems is relatively low cost since there are tools widely used for this purpose, which makes feasible applying software metrics to identify bad smells and to predict faults. Objective: To inspect whether thresholds of object-oriented metrics may be used to aid bad smells detection and fault predictions. Method: To direct this research, we have defined three research questions (RQ), two related to identification of bad smells, and one for identifying fault in software systems. To answer these RQs, we have proposed detection strategies for the bad smells: Large Class, Long Method, Data Class, Feature Envy, and Refused Bequest, based on metrics and their thresholds. To assess the quality of the derived thresholds, we have made two studies. The first one was conducted to evaluate their efficacy on detecting these bad smells on 12 systems. A second study was conducted to investigate for each of the class level software metrics: DIT, LCOM, NOF, NOM, NORM, NSC, NSF, NSM, SIX, and WMC, if the ranges of values determined by thresholds are useful to identify fault in software systems. Results: Both studies confirm that metric thresholds may support the prediction of faults in software and are significantly and effective in the detection of bad smells. Conclusion: The results of this work suggest practical applications of metric thresholds to identify bad smells and predict faults and hence, support software quality assurance activities.Their use may help developers to focus their efforts on classes that tend to fail, thereby minimizing the occurrence of future problems.