Enhancing Project Selection Process Via Machine Learning

Project portfolio management has evolved as businesses attempt to optimize their project selection approaches. In this work, we first investigate the critical determinants of project success in project portfolios. These factors, which are rooted in the typical strategic goals of companies, the project managers' experiences, and the competitive climate, are then used to evaluate new potential projects in terms of their likely success on the dimensions of scope, cost, and time. We collected data on these determinants and the actual success or failures of these projects, which is then used to train various machine learning models, including Decision Trees, Random Forest, XGBoost, and Feedforward Neural Network (FNN), which will then help classify projects based on their potential for success. The dataset includes 284 projects collected from an internal company source under confidentiality constraints. The models aim to predict the likeliness of success of projects competing in a portfolio. ANN outperforms all other models, achieving the highest accuracy in predicting project success, followed by Random Forest and XGBoost, which also performed well but with slightly lower predictive power.