Improving the accuracy of software effort estimation based on multiple regressions by adaptive recursive data partitioning

Abstract

Least squares regression (LSR) has been a popular software effort estimation method in practice; however, an effort estimation model by LSR, a single LSR model, is highly affected by data distribution. Specifically, the model does not provide sound effort estimates in widely scattered data sets because it is very sensitive to influential data points (e.g., outliers), which can distort the model and decrease the estimation accuracy. This implies that it is necessary to develop a data partitioning-based approach as a means to generate multiple LSR models to alleviate the effect of data distribution. Even though clustering-based approaches have been introduced, they have not been sufficiently stable to facilitate accurate effort estimation. In this paper, after we empirically investigate the effect of eliminating outliers on the estimation accuracy of LSR, we propose a new data partitioning-based approach to achieving more accurate and stable effort estimates via LSR. This approach also provides an effort prediction interval that is useful to describe the uncertainty of the estimates. Empirical experiments are performed to evaluate the performance of the proposed approach by comparing with the basic LSR approach and clustering-based approach, based on industrial data sets. The experimental results show that the proposed approach not only improves the accuracy of effort estimation more significantly than that of other approaches, but it also achieves robust and stable results. The proposed approach can help project managers to make accurate and stable effort estimates by alleviating the effect of data distribution that is a major practical issue in software effort estimation.