Software effort estimation using Neuro-fuzzy approach

A successful project is one that is delivered on time, within budget and with the required quality. Accurate software estimation such as cost estimation, quality estimation and risk analysis is a major issue in software project management. A number of estimation models exist for effort prediction. However, there is a need for novel model to obtain more accurate estimations. As Artificial Neural Networks (ANN's) are universal approximators, Neuro-fuzzy system is able to approximate the non-linear function with more precision by formulating the relationship based on its training. In this paper we explore Neuro-fuzzy techniques to design a suitable model to utilize improved estimation of software effort for NASA software projects. Comparative Analysis between Neuro-fuzzy model and the traditional software model(s) such as Halstead, WalstonFelix, Bailey-Basili and Doty models is provided. The evaluation criteria are based upon MMRE (Mean Magnitude of Relative Error) and RMSE (Root mean Square Error). Integration of neural networks, fuzzy logic and algorithmic models into one scheme has resulted in providing robustness to imprecise and uncertain inputs.