A Hybrid Machine Learning and Seasonal Time Series Framework for Variant-Level Monthly Car Sales Forecasting in the Automotive Industry

Abstract

For the automotive sector to improve production planning, inventory control, and strategic decision-making, accurate automobile sales forecasting is crucial. In order to enhance variant level monthly automobile sales prediction, this study suggests a hybrid forecasting framework that blends seasonal time series methods with machine learning models. To find feature-driven sales patterns, the study uses real dealership data and a variety of machine learning approaches, like Random Forest, XGBoost as well as Linear Regression. Simultaneously, seasonal and temporal trends are captured using the SARIMA and Facebook Prophet models. Both strategies
are combined in the suggested hybrid framework to produce accurate and dependable monthly projections for specific car models. RMSE, MAE, and R2 measures are used to assess forecast accuracy. Prophet and Random Forest performed the best out of all the models. In order to provide a clear understanding of model-wise patterns, visualizations that integrate historical and projected data are created. With the help of this variant-level forecasting methodology, Original Equipment Manufacturers (OEMs) and dealerships can make better decisions about marketing and stock allocation. The suggested method helps improve forecasting accuracy and business efficiency in the automobile industry because it is flexible and scalable for practical application.