Group-Specific SVM With Bilevel Programming Methods for Parameter Optimization and Explainable AI in Urban Quality of Life Prediction

The assessment of urban quality of life (UQoL) is a multidimensional challenge involving socioeconomic, environmental, and infrastructure-related variables that, due to changes in the behavior and structures of social issues, requires robust and effective analytical techniques. This study introduces a novel machine learning-based approach incorporating bilevel programming methods to optimize predictive models for UQoL. A key contribution is the development of a Support Vector Machine (SVM) model for group-specific regression, enhanced by a bilevel programming reformulation that optimizes the hyperparameters of the model over groups of data. The bilevel models consider linear and quadratic loss functions. The proposed approach is benchmarked against Linear Regression, Regression Tree, Random Forest, and Gradient Boosting models. The evaluation is conducted using a cross-validation procedure computing the Mean Absolute Error, the Mean Squared Error, and the Mean Absolute Percentage Error as performance metrics. The results indicate that the bilevel quadratic loss SVM model outperforms other models, achieving the lowest prediction error. Furthermore, explainable AI techniques are applied with the usage of Shapley values to analyze feature importance, revealing that own incomes, teacher-to-population ratios, and birth rates are key predictors of UQoL. These findings provide valuable insights for policymakers, supporting data-driven decision-making processes for urban development and resource allocation. This study highlights the potential of AI-driven methodologies in advancing UQoL assessments and urban planning strategies as support systems.