evaluating national energy efficiency using hybrid dea-cross efficiency and machine learning models

Energy efficiency is critical for the attainment of sustainable development, as it optimizes resource utilization and reduces environmental impacts. this study evaluates the energy efficiency of 28 countries from 1995 to 2021 using a hybrid methodology, data envelopment analysis (dea)-cross-efficiency and machine learning models. dea was utilized to compute efficiency scores by analyzing inputs including population and total energy consumption, with output such as total energy production. the scores underwent additional analysis employing six machine learning models: lightgbm, xgboost, knn, random forest, decision tree, and svr. this approach aimed to reveal intricate relationships between the inputs and efficiency ratings, in addition to forecasting future efficiency trends. lightgbm demonstrated outstanding performance, achieving r² = 0.9820, mse = 0.0008, and mae = 0.0155. this performance can be attributed to its capacity to manage large datasets, optimize memory utilization, and implement sophisticated tree-based algorithms for precise predictions. analysis of feature importance indicated that gas and coal production per capita are significant factors influencing energy efficiency. the findings offer policymakers practical insights for optimizing resources and highlight the effectiveness of machine learning in improving conventional efficiency evaluations. in the assessment of the countries, australia and canada exhibited the highest energy efficiency scores, indicative of their proficient resource management and energy policies. these insights provide a framework for other nations to implement comparable strategies aimed at enhancing energy efficiency and fostering sustainable development.