Software Fault-proneness Prediction using Module Severity Metrics

Most of the fault prediction studies have focused on the binary classification models that determine whether the input modules are fault-prone or not. More recently, several studies have shown that severity-based multi-classification models are more useful since they can predict the fault-proneness depending on the severity of the defects in the module. We present new severity-based prediction models using two module severity metrics proposed in our previous study. Those metrics defined to measure the severity of a module are used to define the output values of the prediction model, i.e., the severity-based fault-proneness of a module. Experimental results show that the proposed MS model outperforms previous models in terms of AUC and Accuracy.