A comparison of ARIMA, neural network and a hybrid technique for Debian bug number prediction

A bug in a software application may be a requirement bug, development bug, testing bug or security bug, etc. To prediet the bug numbers accurately is a challenging task. Advance knowledge about bug numbers will help the software managers to take decision on resource allocation and effort investments. The developers will be aware of the number of bugs in advance and can take effective steps to reduce the number of bugs in the new version. The end user can take decision on adopting a particular software application among a variety of applications by knowing the bug growth patterns of the particular software application. The choice of predicting models becomes an important factor for improving the prediction accuracy. This paper provides a combination methodology that combines ARIMA and ANN models for predicting the bug numbers in advance. This method is examined using bug number data for Debian which is publicly available. This paper also gives a comparative analysis of forecasting performance of hybrid ARIMA + ANN, ARIMA and ANN models. Empirical results indicate that an ARIMA-ANN model can improve the prediction accuracy.