A Comparison Study Between Soft Computing and Statistical Regression Techniques for Software Effort Estimation

In this paper, we conduct a comparison between soft computing and statistical regression techniques in terms of a software development estimation regression problem. Our study includes both support vector regression (SVR) and artificial neural network (ANN) as soft computing methodologies on one side, and stepwise multiple linear regression and log linear regression as statistical regression methods on the other side. The experiments are conducted using NASA93 dataset from the well-known PROMISE software repository. Multiple dataset preprocessing steps are performed in order to guarantee confident results including outliers study. We rely on the holdout technique associated with 25 random repetitions with confidence interval calculation within 95% statistical confidence level. Pred(30) evaluation criteria, from literature, is employed to compare between different models. Also, examining different feature combinations, in the case of SVR, shows significant impact on the model precision.