Machine learning-based analysis of urban regeneration and household energy consumption for sustainable energy optimization in deteriorated urban areas

Abstract Optimizing and reducing energy consumption is a central empirical priority for advancing sustainable urban development and mitigating associated environmental risks. Household energy use represents a substantial share of total urban demand, and its dependence on fossil‑based fuels significantly contributes to urban air pollution. This study examines how urban regeneration can reduce residential resource consumption—including water, electricity, and gas—in District 10 of Tehran Municipality. The research is developmental–applied in purpose and descriptive–analytical in design, employing a quantitative modeling framework based on machine‑learning algorithms. The statistical population includes all residential buildings in District 10. Evaluation indicators comprise household water, electricity, and gas consumption, along with 35 additional variables categorized into four dimensions: economic, social–cultural, physical, and environmental. These dimensions were aligned with the variables actually entered into the models to ensure consistency between the conceptual framework and the machine‑learning predictors. Four independent models were tested: Random Forest (RF), Multi‑Layer Perceptron (MLP), Gaussian Process with Radial Basis Function kernel (GP‑RBF), and Gaussian Process with Pearson VII kernel (GP‑PUK). Spatial analyses were conducted in ArcGIS, and statistical preprocessing and classification were performed in Excel. Model comparisons showed that Gaussian Process variants provided the strongest performance overall, though with substantial variation across resource types. For water consumption, the GP‑RBF model achieved a correlation coefficient (CC) of 0.42, RMSE of 0.24, and MAE of 0.19. For electricity consumption, the GP‑PUK model performed best, with CC = 0.99 and RMSE and MAE = 0.001; these values were derived from normalized data and should be interpreted accordingly. For gas consumption, the GP‑RBF model demonstrated the highest accuracy (CC = 0.49), resolving inconsistencies between the Results and Abstract. While the electricity model exhibited strong predictive capability, the moderate correlations for water and gas indicate more limited predictive strength and should be interpreted cautiously. Therefore, the findings suggest that urban regeneration can influence residential resource consumption patterns, and Gaussian Process models—particularly GP‑PUK and GP‑RBF—are effective tools for capturing these relationships. The results provide useful insights for urban planners and policymakers seeking to enhance energy efficiency and resource sustainability within aging urban fabrics.