Fuzzy Cognitive Maps for Software Fault Prediction

Detection of faulty modules in the early stages of the software development life cycle is crucial for the testing procedures. Several software metrics are collected during and end of the development process to represent software modules. By utilizing these collections of module representation, machine learning methods are utilized to predict the fault-prone modules. However, these methods lack interpretability and generalization utilities. In other words, these solutions are highly dependent on the underlying dataset since they aim to discover the hidden relationship between the input to the output in one direction. Addressing these issues, a fuzzy cognitive map has been first proposed to both provide interpretability and to eliminate data dependency for software fault prediction. The proposed cognitive map was learned from experts without the need or dependency on a prior project dataset. In addition to the input-output relations, the relations between inputs were jointly considered. Basing two Mamdani fuzzy inference systems' prediction performance, the proposed map could provide more plausible and accurate decisions in predicting the faulty modules.