Long Short-Term Memory-RNN based model for Multivariate Car Sales Forecasting

Abstract

The paper presents a study of deep learning-based models for forecasting future directions of car sales, and car model preferences. An open-source Kaggle multivariate datasets for many years available for Norway new car sales. They are used for analyzing and predicting. The results based on Autoregressive Integration Moving Average (ARIMA) and Long Short-Term Memory-Recurrent Neural Network (LSTM-RNN) based models are analyzed and used for forecasting future directions. The present study is useful for identifying features of different variants of all used (imported) cars, electric-used cars, and new diesel cars. The implementation results showed reduced Mean Absolute Error (MAE) and Root-Mean-Square Error (RMSE) for LSTM-RNN based time-series forecasting. The study forecasts the rise of green vehicles in the upcoming years in Norway. The performance of a model depends upon the characteristics of the dataset. The results show that LSTM-RNN is thus better than the ARIMAfor the multivariate datasets. The interpretation of these results shows LSTM-RNN based time-series forecasting can be utilized for valuable forecasting in various domains such as credit, insurance, consumer behavior, and medical diagnosis.