An Optimal Neural Network Model for Software Effort Estimation

This paper is concerned with constructing software effort estimation modelbased on artificial neural networks. The model is designed accordingly to improvethe performance of the network that suits to the COCOMO Model. It isproposed to use single layer feed forward neural network to accommodate themodel and its parameters to estimate software development effort. The networkis trained with back propagation learning algorithm and Resilient Back propagationalgorithm (RPROP) by iteratively processing a set of training samples andcomparing the network’s prediction with the actual effort. COCOMO dataset isused to train and to test the network and it was observed that proposed neuralnetwork model improves the estimation accuracy of the model. The test resultsfrom the trained neural network are compared with that of the COCOMO model.By comparing the results of these two models, it is proven that both models(SLANN with BP and SLANN with RPROP) works better than COCOMO andSLANN with RPROP is an optimal neural network model for software effort estimation.SLANN with BP works well only for projects with small size, where asSLANN with RPROP works well for all kinds of projects as the convergencerate of RPROP algorithm is very high. The preliminary results obtained suggestthat the proposed architecture can be replicated for accurately forecasting thesoftware development effort.