A Framework for Early Detection of Agile Software Development Project Failures using Machine Learning Algorithms

Abstract

In the realm of software development, the early identification of project failures is crucial for ensuring project success and minimizing risks. This study developed an early detection framework using various machine learning models to anticipate potential failures. Agile software projects were used for the study. The framework employed a range of machine learning models including decision tree, bagging classifier, weighted bagging classifier, random forest classifier, weighted random forest classifier, decision tree estimator, and bagging estimator. These models are trained and tested using a dataset comprising 13,238 observations from 12 different software companies, each with 15 variables relevant to project performance and outcomes. Initial training of the different models yielded promising results, with performance ranging between 45% to 55% accuracy during testing. Despite attempts to enhance the model's performance, including refinement of features and algorithms, there were no significant improvements observed. The evaluation results highlight the need for further refinement and optimization of the models used in the framework. In conclusion, while the decision tree classifier, bagging classifier, and random forest exhibited outstanding performance in the training results, the overall evaluation suggests that more work is required to improve the effectiveness of the early detection framework for Agile software project failures. Further research and refinement of the models are necessary to enhance accuracy and reliability in identifying potential project failures early in the Agile software development lifecycle.