Determination of energy efficiency based on machine learning and efficiency assessment methods

Effectively measuring energy efficiency (EE) and gaining a comprehensive understanding of its status are essential toward enhancing overall EE. Traditional EE evaluation methods typically rely on relative and linear efficiency frontiers, which are inadequate for handling the scenarios of extremes, outliers, and the incorporation of new decision-making units (DMUs), often resulting in unreliable results. Aiming to address this issue, this study applies machine learning algorithms to optimize conventional efficiency evaluation methods by constructing an absolute and smooth efficiency frontier. This enhanced method is then used to perform precise EE measurements across 30 provinces in China. The empirical analysis yields several key insights. First, a notable difference exists between the EE values calculated by the proposed model and those calculated by traditional models, with the proposed model offering higher efficiency values that more accurately reflect the developmental characteristics of DMUs. Second, EE across China and its regions exhibited an upward trend from 2000 to 2023. Third, output-based analysis reveals that the northeastern region exhibits the greatest deviation from the ideal level in terms of undesirable output, while the western region shows the greatest discrepancy in terms of desirable output. Based on the empirical analysis, target policy recommendations are proposed to further improve EE in China.