Title: A comparative study of automobile sales forecasting with ARIMA, SARIMA and deep learning LSTM model

Authors: Sharath Kariya Shetty; Rajesh Buktar

Addresses: Mechanical Engineering Department, Sardar Patel College of Engineering, Bhavan's Campus, Mumbai, India ' Mechanical Engineering Department, Sardar Patel College of Engineering, Bhavan's Campus, Mumbai, India

Abstract: In deciding production-plan, material-inventory, scheduling, etcetera of an automobile industry, the accuracy of forecasting techniques plays a very important role. The quantitative forecasting techniques in the automobile industry, conventional methods like auto-regressive, ARIMA, and seasonal-ARIMA are not accurate enough to extract all the information hidden in the time series data. With the recent developments in machine-learning and deep learning, RNN and LSTM can produce an impressive result out of time series data by dealing with the nonlinearity and complexity of the data. In this study, we have compared the prediction accuracy among the ARIMA, SARIMA, and LSTM techniques for an Indian automobile company. The deep learning model LSTM outperforms by 92% when compared with ARIMA and by 42.5% when compared with SARIMA techniques. Moreover, it is noticed that the accuracy of deep learning can be improved by tuning hyperparameters such as learning rate, neuron number in a layer, and choosing the correct weight initialiser for the data.

Keywords: automobile sales; forecasting; ARIMA; SARIMA; deep learning; LSTM.

DOI: 10.1504/IJAOM.2022.127864

International Journal of Advanced Operations Management, 2022 Vol.14 No.4, pp.366 - 387

Received: 11 Sep 2021
Accepted: 27 Jan 2022
Published online: 20 Dec 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article