This study aims to bridge the gap between the high predictive accuracy of machine learning models and the transparency requirements of public policies for sustainable mobility by analyzing technological adoption within the national vehicle fleet of Ecuador. To this end, a methodological framework grounded in Design Science Research (DSR) and CRISP-DM was applied to a real administrative dataset of 482,754 vehicle registration records from the Ecuadorian Internal Revenue Service (SRI, 2025), comparing the performance and interpretability of two state-of-the-art gradient boosting algorithms, XGBoost and LightGBM, for multiclass classification of internal combustion engine (ICE), hybrid, and electric vehicles (EV) under severe class imbalance. The results demonstrate that both models achieve near-perfect predictive performance, with a consolidated Macro F1-score of 0.987, confirming their robustness and suitability for large-scale, policy-relevant administrative data even when electric vehicles represent only 1.3\% of the observed fleet. Beyond global performance metrics, the integration of explainable artificial intelligence through SHAP values reveals that fiscal appraisal value and engine capacity are the primary determinants of EV adoption, while territorial factors exhibit greater influence in the case of hybrid vehicles, highlighting qualitatively distinct adoption mechanisms across technologies. These findings show that advanced ``black-box'' models can be transformed into auditable and interpretable analytical tools capable of linking predictions to concrete economic, technical, and spatial variables relevant for decision-making. The study concludes that XAI-enabled gradient boosting provides a rigorous and transparent framework to support the design of targeted fiscal incentives, fleet electrification strategies, and evidence-based transport decarbonization policies in emerging economies, where data-driven governance is critical for accelerating the transition toward sustainable mobility.