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

Area of Science:

• Environmental Science
• Automotive Engineering
• Data Science

Background:

• Electric vehicles (EVs) are promoted to mitigate climate change and reduce fossil fuel dependence.
• Accurate EV sales forecasting is vital for manufacturers, policymakers, and energy providers.
• Data quality and modeling techniques critically influence prediction accuracy.

Purpose of the Study:

• To develop and evaluate advanced machine learning models for predicting monthly electric vehicle sales in the USA.
• To introduce a novel hybrid deep learning architecture for enhanced forecasting performance.
• To compare the proposed model against established methods using comprehensive evaluation metrics.

Main Methods:

• Utilized monthly sales and registration data for 357 new vehicles in the USA (2014-2020).
• Employed Long Short-Term Memory (LSTM) and Convolutional LSTM (ConvLSTM) models.
• Developed a novel hybrid LSTM model incorporating two-dimensional Attention and Residual networks, built using Automated Machine Learning (AutoML).

Main Results:

• The proposed hybrid LSTM model demonstrated superior performance over standard LSTM and ConvLSTM models.
• Evaluation metrics included Mean Absolute Percentage Error, Normalized Root Mean Square Error, R-square, slope, and intercept.
• The hybrid model achieved a Mean Absolute Error of 3.5% in predicting EV market share.

Conclusions:

• The novel hybrid LSTM model offers a significant advancement in electric vehicle sales forecasting.
• Automated Machine Learning frameworks enhance the development and performance of deep learning prediction models.
• Accurate EV sales predictions support strategic planning for a sustainable automotive future.