A New Weighted Ensemble Model to Improve the Performance of Software Project Failure Prediction

Abstract

The development of a software project is frequently influenced by various risk factors that can lead to project failure. Predicting potential software project failures early can aid organizations in making decisions regarding possible solutions and improvements. This paper proposes a software project failure prediction model based on a weighted ensemble learning approach. The proposed model aims to determine the failure probability as well as the expected project outcome (Success/Failure). Various ensemble approaches, such as simple majority voting, can be employed in predicting software project failure. However, in majority voting algorithms, all base models have the same weights, resulting in an equal effect on the final prediction result, regardless of their predictive abilities. Our proposed algorithm assigns higher weights to base models that demonstrate a greater ability to correctly predict more challenging data instances. The proposed model is developed based on a dataset gathered from real previous software project reports, comprising both successful and failed projects, to provide evidence supporting the predictive model's capabilities and to obtain high-confidence results. The performance of the developed model is comprehensively assessed through various measures, revealing its superiority in predicting software project failures compared to both simple majority voting and individual models. This research suggests that the proposed model can be integrated into the software system development process, spanning requirement analysis, planning, design, and implementation phases, to evaluate the project's status and identify potential risks.