A Non-Linear Technique for Effective Software Effort Estimation using Multi-Layer Perceptrons

An effective predictor for software effort estimation is highly desirable for the successful delivery of product in time and within the budget. Both either the underestimation or the overestimation endangers the entire business and market value too. Number of studies has been made in this direction but no univocal conclusion has been attained. It is still an open problem. In this work, a non-linear technique is implemented for the effort estimation using Multi-Layer Perceptrons with back propagation algorithm. The dataset used for this work is Maxwell Dataset. Then the comparison is made with the linear regression technique. The result shows that nonlinear technique is much better than linear technique with the value of 0.13 for MdMRE whereas the linear technique results the value of 0.36 for MdMRE. The results show that attribute selection also affects the overall performance.