ABSTRACT

A multiple linear regression and ARIMA hybrid model is proposed for new bug prediction depending upon resolved bugs and other available parameters of the open source software bug report. Analysis of last five year bug report data of a open source software �worldcontrol� is done to identify the trends followed by various parameters. Bug report data has been categorized on monthly basis and forecast is also on monthly basis. Model accounts for the parameters such as resolved, assigned, reopened, closed and verified bugs respectively. Real time monthly data of these parameters from 2003 to 2007 is taken for multiple regression then hybrid model does monthly forecast for 2008. Model is basically hybrid of linear regression and ARIMA(p,0,p) where p = 1,2,3. Results show that monthly forecast of new bugs considering five predefined factors is far more accurate by hybrid model than just time series ARIMA forecast of new bugs. Hybrid of linear regression and ARIMA (3,0,3) gave best results.