Generating Explanations for AI-Powered Delay Prediction in Software Projects

A project failure can be attributed to complex negative factors that can deviate project progress from the original schedules, and one of the root causes can be a delay. Hence, the early detection of a delay sign can be a critical component for the success of a project. One approach that contributes to solving the problem can be the development of prediction models, and machine learning methods can be a promising approach due to the recent success in other areas. Therefore, we introduce an AI-based novel approach using an explainable graph neural network that elucidates the causes of a delay without compromising its prediction performance. Three experimental results demonstrate that (1) our model can predict the delay with 4% higher accuracy on average, (2) our model returns a stable result by providing a similar prediction performance with a similar explanation when the same prediction tasks are given, and (3) the generated explanations can provide actual reasons for the delay prediction given that the optimal threshold is used. These points can provide a more supportive delay prediction system to users, reducing the failure of projects in terms of time control.