Extreme learning machine for software development effort estimation of small programs

During the last few decades software effort estimation has got the attention of many software engineering researchers both in academia and industry to develop new models. Recently Extreme Learning Machine (ELM) is being applied to many problems where feed forward neural networks are used. It has not been applied for small projects. The performance of ELM is compared with the Linear Least Squares Regression (LSR). The effect of using one or two independent variables is evaluated. The results of the experiments show that ELM is an alternative to LSR and one independent variable can be used for estimating effort without sacrificing the accuracy.