摘要 : The software fault prediction (SFP) literature has shown an immense growth of the research studies involving the artificial neural network (ANN) based fault prediction models. However, the default gradient descent back propagation... 展开 The software fault prediction (SFP) literature has shown an immense growth of the research studies involving the artificial neural network (ANN) based fault prediction models. However, the default gradient descent back propagation neural networks (BPNNs) have a high risk of getting stuck in the local minima of the search space. A class of nature inspired computing methods overcomes this disadvantage of BPNNs and has helped ANNs to evolve into a class of adaptive ANN. In this work, we propose a hybrid SFP model built using firefly algorithm (FA) and artificial neural network (ANN), along with an empirical comparison with GA and PSO based evolutionary methods in optimising the connection weights of ANN. Seven different datasets were involved and MSE and the confusion matrix parameters were used for performance evaluation. The results have shown that FA-ANN model has performed better than the genetic and particle swarm optimised ANN fault prediction models. 收起