Early Criticality Prediction Model Using Unsupervised Neural Network

In this paper, we classify criticality prediction models into four types according to the input metric forms and the necessity of the past project data. Many prediction models for identifying fault-prone modules using complexity metrics have been suggested. But most of them are VI-SL models that need training data set. Unfortunately very few organizations have their own training data. To solve this problem, we build a new VI-UL model, KSM, based on Kohonen SOM neural networks. We compare and evaluate KSM and a well-known VI-SL model, BPM, considering internal characteristics, utilization cost and accuracy of prediction. As a result, we show that KSM has similar performance with the VI-SL model.