Improving demand forecasting by implementing machine learning in poultry production company

Abstract

Debido al alto porcentaje de error presente en el pronóstico de la demanda de una empresa de producción avícola; lo que desencadena una pérdida monetaria mensual provocada por el método manual que utiliza la empresa para el pronóstico del plan de ventas, es por ello que se buscará reducir este porcentaje de error mediante el uso de la herramienta Machine Learning, a través de la cual se utilizará la base de datos de ventas proporcionada por la empresa que se utilizó para entrenar y utilizando el método de series de tiempo, se podrá predecir el último trimestre del 2022. Finalmente, se compararon los resultados obtenidos por el modelo de Machine Learning con la venta real de la empresa y su pronóstico. Se logro como resultado un FB de 2,44% y un FA de 97,56%, reduciendo así el error que manejaba la empresa.

The use of manual methods to forecast demand in perishable food companies is generally subject to the variability of internal and external factors in the company, causing excess inventories and significant monetary losses, so it is relevant to carry out this research with the objective of to demonstrate that by implementing Machine Learning it is possible to improve the accuracy of the demand forecast. A case study in a company in the poultry sector in Peru, forecasting the last quarter of 2022, based on a real sales database and applying the time series method, comparing the results of the Machine Learning model, and obtaining as a result in a model with high Forecast Accuracy (FA) of 97.56% and a high Forecast Bias (FB) of 2.44%. The research is an important contribution to knowledge, demonstrating that Machine Learning is an ideal tool to project the demand for perishable food products, ideal for its application in various fields, such as loss reduction control, preventive maintenance of machines and control of supplies such as water and energy, among others.