Software Cost Estimation Method Based on Weighted Partial Least Squares Regression
Jiang Guo-pinga · Jisuanji gongcheng · 2012
For the excessiveness of operating factors and multi-correlation of variables in software cost estimation,this paer proposes a software cost estimation method based on Partial Least Squares Regression(PLSR).After the variables are weighted,an integrated index named analogue deviation is defined to describe the approximation of data samples.Then an adaptive weight is assigned to sample according to the approximation,and the optimal partial least squares latent variables and weight parameters are calculated by traversal searching.Experimental results show that the prediction error is reduced by 73.61% and 32.34% than multiple linear regression and PLSR respectively.