Using machine learning methods to predict electric vehicles penetration in the automotive market

Shahriar Afandizadeh; Diyako Sharifi; Navid Kalantari; Hamid Mirzahossein

doi:10.1038/s41598-023-35366-3

Using machine learning methods to predict electric vehicles penetration in the automotive market

Sci Rep. 2023 May 23;13(1):8345. doi: 10.1038/s41598-023-35366-3.

Authors

Shahriar Afandizadeh¹, Diyako Sharifi², Navid Kalantari³, Hamid Mirzahossein⁴

Affiliations

¹ Department of Transportation, School of Civil Engineering, Iran University of Science and Technology, Tehran, Iran. zargari@iust.ac.ir.
² Department of Transportation, School of Civil Engineering, Iran University of Science and Technology, Tehran, Iran.
³ AECOM, Glen Allen, VA, USA.
⁴ Department of Civil-Transportation Planning, Faculty of Technical and Engineering, Imam Khomeini International University, Qazvin, Iran.

Abstract

Electric vehicles (EVs) have been introduced as an alternative to gasoline and diesel cars to reduce greenhouse gas emissions, optimize fossil fuel use, and protect the environment. Predicting EV sales is momentous for stakeholders, including car manufacturers, policymakers, and fuel suppliers. The data used in the modeling process significantly affects the prediction model's quality. This research's primary dataset contains monthly sales and registrations of 357 new vehicles in the United States of America from 2014 to 2020. In addition to this data, several web crawlers were used to gather the required information. Vehicles sale were predicted using long short-term memory (LSTM) and Convolutional LSTM (ConvLSTM) models. To enhance LSTM performance, the hybrid model with a new structure called "Hybrid LSTM with two-dimensional Attention and Residual network" has been proposed. Also, all three models are built as Automated Machine Learning models to improve the modeling process. The proposed hybrid model performs better than the other models based on the same evaluation units, including Mean Absolute Percentage Error, Normalized Root Mean Square Error, R-square, slope, and intercept of fitted linear regressions. The proposed hybrid model has been able to predict the share of EVs with an acceptable Mean Absolute Error of 3.5%.