Predicting Future Electric Vehicle (EV) Sales: A Time Series Forecasting Approach Using Historical EV Sales Data

Abstract

Accurate forecasting of Electric Vehicle (EV) sales is essential for supporting strategic decisions by policymakers, manufacturers, and investors amid the global shift toward sustainable transportation. This study compares the performance of two time series models, AutoRegressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) using historical EV sales data from 2010 to 2023. The ARIMA model, which is suited for linear trend projection, forecasts continued exponential growth, estimating sales to surpass 103 million units by 2025. In contrast, the LSTM model, known for capturing non-linear and complex patterns, projects a more moderate trend, with sales peaking at around 11.5 million units in 2022 before gradually declining. Evaluation using Mean Squared Error (MSE) shows that LSTM significantly outperforms ARIMA, achieving a lower error value (2.23 × 10¹⁴ vs. 4.44 × 10¹⁵), indicating superior predictive accuracy. These results suggest that while ARIMA may be effective for short-term forecasting in stable markets, it can lead to overestimations in more dynamic environments. LSTM, with its ability to learn complex temporal dependencies, presents a more flexible and realistic tool for long-term planning in the evolving EV sector. The study contributes methodologically by offering a comparative analysis of two popular forecasting techniques and practically by guiding stakeholders on model selection. However, it is limited by its reliance on historical data and exclusion of external variables such as energy prices or policy changes. Future work should incorporate hybrid models and multi-source data to enhance forecasting robustness in the fast-changing EV market