根据序列变化率预测软件阶段成本

Predicting Software Stage Effort with Sequence Changing Ratio

针对软件阶段成本因少数据、不确定性使得用现有方法(如回归)难以预测的问题,文中提出一种新颖的预测方法,该方法从项目已完成阶段的成本序列中,通过变换得到反映序列变化快慢的"变化率",并用机器学习方法从历史项目中学习得到变化率阈值,然后用不同的灰色模型进行预测.在10个现实世界软件工程数据集上的实验结果表明,该方法平均预测误差比线性回归方法低20%～80%,显示出较大的潜力.

Software stage effort has the features of data starvation and uncertainty. It is difficult to use the current methods （e. g. regression） to make predictions. This paper proposes a novel prediction method, which gets the effort sequence changing feature--＂changing ratio＂ from the completed stage effort sequences, and gets the ＂changing ratio threshold＂ from historical projects by machine learning methods, then uses grey models to make predictions. The experimental results on 10 real world software engineering datasets show that, compared with linear regression method, the prediction accuracy of the proposed method has been improved by 20%-80%. This is very encouraging and indicates that the method has considerable potential.