Hybrid Neuro-Fuzzy Systems for Software Development Effort Estimation

The major prevailing challenges for Software Projects are Software Estimations like cost estimation, effort estimation, quality estimation and risk analysis. Though there are several algorithmic cost estimation models in practice, each model has its own pros and cons for estimation. There is still a need to find a model that gives accurate estimates. This paper is an attempt to experiment different types of Neuro-Fuzzy Models. Using the types of Neuro-Fuzzy Models for software effort prediction is a relatively unexplored area. Two case studies are used for this purpose. The first is based on NASA-93 dataset and the other is based on Maxwell-62 dataset. The case studies were analyzed using six different criterions like Variance Accounted For (VAF), Mean Absolute Relative Error (MARE), Variance Absolute Relative Error (VARE), Mean Balance Relative Error (Mean BRE), Mean Magnitude Relative Error (MMRE) and Prediction. From the results and from reasoning, it is concluded that Type BCompensation Neuro-Fuzzy Model with more fuzzy rules is best suitable for cases in which the datapoints are more linear. Type J Neuro-Fuzzy Model with more fuzzy rules is best suitable for cases in which the datapoints are not linear. Keywords-Effort Estimation, Fuzzy Logic, Neural Nets, Neuro-Fuzzy Models, NASA-93 Dataset, Maxwell-62 Dataset.