Modeling Automobile Sales in Turkiye with Regression-Based Machine Learning Algorithms

Abstract

The automobile sector is the locomotive of industrialized countries. The employment opportunities it creates are of great value because of its interconnectedness with other industries and the value it adds. Demand forecasting studies in such an important sector are one of the main drivers for the provision of raw materials and services needed in the future. In this study, 10 independent variables are used that directly or indirectly affect the level of car sales, which is our dependent variable. These variables are gross domestic product, real sector confidence index, capital expenditures, household consumption expenditures, inflation rate, consumer confidence index, percentage of one-year term deposits, and oil barrel, gold, and dollar prices. The dataset used consists of annual data between 2000 and 2021. To examine the sales forecast model, two variables that affect minimum sales are first extracted from the model using the least squares method. Linear Regression, Decision Tree, Random Forest, Ridge, AdaBoost, Elastic-net, and Lasso Regression algorithms are applied to build a predictive model with these variables. The Mean Squared Error (MSE), Mean Absolute Error (MAE), and coefficient of determination (R 2 ) are used to compare the performance of the predictive models. This study proposes an approach for sectors affected directly or indirectly by automotive sales to gain foresight on this issue.