A Novel Model for Risk Estimation in Software Projects Using Artificial Neural Network

Abstract

Software projects generally involve more risks due to unexpected negative results. Therefore, the risks encountered in software projects should be detected and analyzed on time, and effective precautions should be taken in order to complete the projects successfully. The aim of this study was to estimate the deviations that may occur in the software project outputs according to risk factors by using artificial neural networks (ANNs). Thus we aimed to minimize loses that may occur in project processes with the developed model. Firstly, a comprehensive and effective list of risk factors was created. Later, a checklist form was prepared for Team Members and Managers. The data collected include general project data and risk factors, and these are the inputs of the model. The outputs of the model are the deviations in the project outputs. MATLAB package program was utilized to develop the model. The performance of the model was measured according to Regression Values and Mean-Squared Error. The model obtained has forty-five inputs, one hidden layer that has fifteen neurons, and five outputs (45-15-5). In addition, the training-R, testing-R, and MSE values of the model were found as 0.9978, 0.9935, and 0.001, respectively. It is seen that the estimation results obtained with the model using the real project data coincide with the actual results largely and the error rates were also very low (close to zero). The experimental results clearly revealed that model performance is high, and it is very effective to use ANNs in risk estimation processes for software projects.