Estimación de tiempos de carga de transportes mediante modelos Machine Learning para una planta de distribución de productos de nutrición de cultivo

Abstract

This project aims to implement an interactive visualization tool based on machine learning models to support logistical planning and improve loading efficiency in a crop nutrition production plant. The CRISP–DM methodology was applied to structure the process, including business understanding, exploratory analysis, data preparation, predictive modeling, and deployment within Power BI. Supervised regression models (Linear Regression, Random Forest, and XGBoost) were trained, with Random Forest achieving the best performance. The model reduced the average planning error from 14.02 to 7.38 minutes, representing a 47.3% improvement compared to the current manual estimation method used by the plant. The visualization tool enables planners to view predicted loading times, compare them with actual values, and identify operational deviations through indicators such as OTIF. The economic assessment shows operational savings of approximately USD 1,200 to USD 1,600 per month, equivalent to around USD 18,000 annually, excluding indirect benefits such as reduced idle time and improved process stability. The study demonstrates the feasibility of integrating machine learning techniques with business intelligence tools in real logistics environments, providing a methodological framework that can be replicated in similar industrial operations. Keywords: logistics, machine learning, Random Forest, CRISP–DM, data visualization, Power BI, loading time estimation.