An empirical analysis of linear adaptation techniques for case-based prediction

Abstract

This paper is an empirical investigation into the effectiveness of linear scaling adaptation for case-based software project effort prediction. We compare two variants of a linear size adjustment technique and (as a baseline) a simple k-NN approach. These techniques are applied to the data sets after feature subset optimisation. The three data sets used in the study range from small (less than 20 cases) through medium (approximately 80 cases) to large (approximately 400 cases). These are typical sizes for this problem domain. Our results show that the linear scaling techniques studied, result in statistically significant improvements to predictions. The size of these improvements is typically about 10% which is certainly of value for a problem domain such as project prediction. The results, however, include a number of extreme outliers which might be problematic. Additional analysis of the results suggests that these adaptation algorithms might potentially be refined to cope better with the outlier problem.