Analytical tool prototype for monitoring road infrastructure projects in Colombia incorporating machine learning

Abstract

This article presents the development of an analytical tool prototype designed for the control of road infrastructure projects in Colombia, incorporating machine learning techniques as both a pedagogical resource and an innovation in project management. Its purpose is to strengthen decision-making and training processes in both academic and professional settings through predictive models that estimate project duration and costs, optimize resources, and generate automatic technical recommendations. The methodology followed the CRISP-DM approach, starting with the collection and cleaning of historical data from the Infrastructure Project Manager (GPI), and the development of four models: LightGBM for duration prediction, K-Means for resource optimization, linear regression for cost estimation, and Random Forest for recommendations. These models were integrated into an interactive interface that enables real-time use, fostering applied learning and reflective analysis. The results show high predictive accuracy and effective classification into three efficiency levels. It is concluded that incorporating data-driven analytical tools in educational and professional contexts not only improves project control and planning but also fosters critical competencies in data analysis, risk mitigation, and strategic decision-making in complex scenarios.