Abstract:

Shrinkage in perishable products represents a structural problem in food supply chains, with rates that can exceed 10–15% of the volume handled during distribution and storage stages. The present study designs and evaluates a prototype of an intelligent system that integrates machine learning, Monte Carlo simulation, and multi-objective optimization to reduce shrinkage in the Peruvian logistics context. A synthetic database was constructed with representative operational and environmental variables, including temperature, humidity, inventory level, demand, and storage time, on which a hybrid model combining Random Forest Regression with artificial neural networks was trained. The model obtained a coefficient of determination R² = 0.8410 and a mean absolute percentage error (MAPE) = 16.65%, confirming its predictive capability. The Monte Carlo simulation with 3,000 scenarios identified a critical shrinkage threshold of 11.2788%, classifying 20% of the scenarios as high operational risk events. The comparison with a conventional system estimated an average reduction in shrinkage of 24.46%, validated through a 95% confidence interval that confirmed the absence of systematic bias. The results demonstrate that the integration of artificial intelligence and stochastic simulation transforms perishable inventory management from a reactive approach toward a preventive one, with direct applications in Peruvian companies in the food sector.