Forecasting electric two-wheeler demand in the Indian supply chain using machine learning models

Abstract

India’s electric two-wheeler (E2W) market is growing rapidly, driven by policies, fuel prices, and EV infrastructure. Accurate demand forecasting is vital for effective planning and supply chain decisions. However, most existing studies overlook E2W-specific time-series forecasting using interpretable machine learning (ML) models. This study addresses the research gap by developing a forecasting framework that uses advanced ML models to predict monthly 2WN registrations in India. The objective is to find a model that minimizes prediction error and offers strong explanatory power. Monthly registration data from the VAHAN portal (2018–2024) was used. After cleaning, the dataset was enhanced with features like lag values, rolling averages, and growth indicators. Three ML models were implemented and evaluated: Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosted Trees (GBT). GBT achieving the best performance (RMSE: 3103.26; R²: 0.902) after hyperparameter tuning. RF achieved moderate performance, while SVM struggled with non-linear and sequential patterns. The engineered features and multivariate time-series structure significantly enhanced model accuracy and generalizability. The proposed approach enables reliable six-month-ahead forecasts for E2W registrations. This has high real-world value for manufacturers, policymakers, and planners seeking to anticipate demand, manage supply chains, and deploy EV infrastructure effectively. By offering a precise, interpretable, and data-driven forecasting tool, the study contributes to strategic planning in India’s transition to sustainable mobility.