Predicting Project Effort Intelligently in early Stages by Applying Genetic Algorithms with Neural Networks
In the early stages of a software development project, estimating the amount of effort is one of the most important project management concerns. This study has successfully produced global optimal reduced models intelligently predicting software cost estimation by employing neural networks with back-propagation learning algorithms combined with genetic algorithms (GA-NN) to determine the most significant explanatory variables among the 16 COCOMO cost drivers. The performance of the full model of GA-NN is much superior to that of the COCOMO, whilst the predicting performance of its global optimal reduced model is also comparable to that of the COCOMO in terms of MMRE and PRED (25). The optimal reduced models and their found significant factors can offer powerful supports for the project managers to make right decisions in the early stages of the projects.