FORECAST MODELS AND ARTIFICIAL INTELLIGENCE APPLICATION TO EVALUATE DEMAND IN THE AUTOMOTIVE LIGHTING SYSTEMS SEGMENT

Abstract

The Market competition encourages the companies to seek innovation in order to stand out, bringing value to the segment they operate in. This strong competition creates a need for organizations to look for ways to reduce costs and streamline processes; the accuracy on the demand forecasting is an essential factor to improve productivity, stock management and lead time reduction, contributing to the company’s results. This paper’s goal is to use mathematical models from time series and artificial intelligence to check which method is the most accurate in forecasting the demand of a lighting automotive company. The Artificial Neural Networks (RNA) method was optimized through settings considering the number of neurons and different training algorithms to find the most accurate models. Through the application of the MAPE and MAE forecasting errors, the conclusion was that some RNA settings are the most accurate methods to make predictions of the analyzed products, and that, in average, the optimized RNA forecasting errors are 3,25 times (for MAPE) and 4 times (for MAE) lower than the actual company’s method.