Discriminating risky software project using neural networks

• This study presents a predictive model to effectively predict riskiness of a project.
• Artificial neural networks can be used to learn the complex patterns of the OMRON dataset.
• Performance is evaluated and shown to be more advantageous than the logistic regression model.
• It provides better accuracy and sensitivity.

Early and accurate discrimination of risky software projects is critical to project success. Researchers have proposed many predictive approaches based on traditional modeling techniques, but the high misclassification rate of risky projects is common. To overcome this problem, this study proposes a typical three-layered neural network (NN) architecture with a back propagation algorithm that can learn the complex patterns of the OMRON dataset. This study uses four accuracy evaluation criteria and two performance charts to objectively quantify and visually illustrate the performance of the proposed approach. Experimental results indicate that the NN approach is useful for predicting whether a project is risky. Specifically, this approach improves accuracy and sensitivity by more than 12.5% and 33.3%, respectively, compared to a logistic regression model developed from the same database. These results imply that the proposed approach can be used for early planning of limited project/organization resources and appropriate action for risky projects that are likely to cause schedule slippage and cost overload.