Time series prediction of debian bug data using autoregressive neural network

Predicting the increasing or decreasing bug numbers is an important factor that affect the decision making process of the software managers. Software managers can make timely decisions, such as effort investment and allocation of resources by predicting the bug number of a software system accurately. The objective of this paper is to model the bug number data per month as time series and, and analyzing the time series using Artificial Neural Network(ANN). A Nonlinear autoregressive model(NAR) with embedded delay and feedback loop is used for time series prediction of debian bug data. This paper gives a complete neural network approach to bug number prediction. A comparison of five most popular neural net Training algorithms is given in this paper. The results shows a substantial improvement in performance of LevenbergMarquardt algorithm with Bayesian Regularization than other training algorithm. The results are confirmed on bug data extracted from bug Ultimate Debian Database(UDD) which is publicly available.