Experimental validation of inheritance metrics' impact on software fault prediction

Abstract

Software faults can cause trivial annoyance to catastrophic failures. Recent work in software fault prediction (SFP) advocates the need for predicting faults before deployment to aid testing process. Object-oriented programming is complex while comparing it with procedural languages having multiple dimensions wherein inheritance is an important aspect. In this paper, we aim to investigate how much inheritance metrics assist in predicting software fault proneness. We first select the Chidamber and Kemerer (CK) metrics, most accepted metric suite for predicting software faults and inheritance metrics. We use 65 publicly available base datasets having CK metrics and some other inheritance metrics to evaluate the impact of inheritance on SFP. We split each dataset into further two datasets: inheritance with CK and CK without inheritance for comparison of results. An artificial neural network (ANN) is used for model building, and accuracy, recall, precision, F1 measures, and true negative rate (TNR) are used for measuring performance. Comparison is made and the results show an acceptable contribution of inheritance metrics in SFP. The testing community can safely use inheritance metrics in predicting software faults. Moreover, high inheritance is not desirable, as this can potentially lead to software faults.