Uso de técnicas de previsão de demanda para redução de sobras de alimentos em um restaurante universitário

Abstract

University restaurants ensure students’ access to adequate meals, encouraging student retention at the university. However, the efficient management of food services in public higher education institutions represents a significant challenge, especially in view of the need to meet the growing demand of diners while ensuring quality and optimizing resources. Among the various factors that influence waste, the difficulty in accurately predicting the number of diners stands out, since demand is affected by academic, behavioral, and seasonal variables. In this context, demand-forecasting techniques based on time series emerge as tools to improve planning and reduce the surplus generated daily in Food and Nutrition Units. Therefore, this study aims to analyze the number of diners at the University Restaurant of the Federal University of Ceará – Crateús Campus, using time-series-based demand-forecasting techniques in order to reduce the rate of food leftovers and, consequently, food waste. The research was characterized as applied, descriptive, quantitative, and as a case study. Historical data on consumption and leftover percentages from the semesters 2023.1, 2023.2, 2024.1, and 2024.2 were used. The predictive techniques evaluated were moving average, weighted moving average, simple exponential smoothing, double exponential smoothing (Holt), triple exponential smoothing (Holt-Winters), and Box-Jenkins. Model selection considered the error metrics MAD (Mean Absolute Deviation) and MAPE (Mean Absolute Percentage Error). The results indicated that the additive Holt-Winters method showed the best predictive performance, with mean MAPE values around 11.9% and an approximate MAD of 32 meals, outperforming the other methods tested. This technique was applied for seven weeks during the 2025.2 semester, with forecasts adjusted based on operational information provided by the university restaurant management. Statistical analysis using Student’s t-test and Pearson’s correlation showed that, at lunch, the reduction in the average percentage of leftovers was not statistically significant. On the other hand, at dinner, a significant reduction in leftovers was observed (p < 0.05), as well as a significant correlation between the adjusted MAPE and the volume of leftovers, indicating greater sensitivity of this meal period to forecasting accuracy. The application of demand-forecasting methods, combined with managerial adjustments, proved effective in reducing waste and contributing to improved restaurant planning. As a practical contribution, the study resulted in the development of a systematized technical procedure for the weekly application of the Holt-Winters technique, providing support for production planning and leftover control in the university restaurant.