Abstract
The non-homogeneous Poisson process (NHPP) is a commonly used method for developing software reliability growth models (SRGM). These models are utilized for several significant predictions, including the remaining number of faults, defect rate, time to next defect, and reliability. However, SRGMs cannot be applied for long-term prediction using a limited amount of data. To overcome this limitation, an autoregressive model incorporating the windowing (ARMIW) technique is proposed for improved prediction of software defects. The ARMIW is illustrated by applying it to several software defect times datasets. The long-term predictive capability of the ARMIW is also illustrated by fitting the model over several fit-test ratios. The results demonstrate that the proposed modeling technique estimates the model parameters with higher precision in comparison to the traditional autoregressive (AR) models as well as the NHPP SRGMs and improves the time-series prediction accuracy. Results suggest that the autoregressive model incorporating windows improves the prediction of the defect discovery process significantly and makes better long-term predictions even using a small amount of available data. The advantage of incorporating windows in auto-regression is that they provide a better model fit. The AWMIW exhibits an 11-fold improvement in defect time prediction over the traditional AR model and a fourfold improvement compared to NHPP SRGMs.