Quantum-inspired scenario-adaptive differential privacy with hybrid attention–residual deep learning for utility-aware urban location analytics

Abstract Location-based services (LBS) increasingly rely on trajectory data, creating a persistent challenge in balancing user privacy and data utility. Conventional differential privacy (DP) mechanisms often employ fixed noise distributions and global privacy budgets, limiting their ability to adapt to heterogeneous urban environments. To address this limitation, we propose a Quantum-Inspired Scenario-Adaptive Differential Privacy framework that combines k-nearest-neighbor (kNN)-based adaptive privacy budgeting with structured quantum-inspired noise generation and a Hybrid Attention–Residual Deep Neural Network for mechanism selection. The framework introduces a five-dimensional urban morphology descriptor comprising Uniformity, Directional Bias, Spatial Correlation, Density Gradient, and Clustering, and jointly learns privacy-mechanism selection and urban-pattern classification through a dual-task architecture. The Hybrid DNN achieved a mean mechanism-classification accuracy of 97.35% ± 0.24 across ten random seeds, performing competitively with XGBoost (97.33% ± 0.20), a Tabular Transformer (97.26% ± 0.31), and a GraphSAGE-style Graph Neural Network (97.24% ± 0.28). Ablation analysis under data-scarce noisy conditions identified dual-task learning as the most influential architectural component, yielding a statistically significant accuracy improvement of 0.33 percentage points Wilcoxon p = 0.035. Across the privacy benchmark, quantum-inspired Laplace mechanisms reduced mean absolute error by up to 33.19%, utility loss by up to 41.36%, privacy loss by up to 38.48%, and geo-indistinguishability error by up to 51.61% relative to their classical counterparts. Quantum Anisotropic Gaussian achieved the largest utility gains, reducing distortion metrics by 72–78%, while Quantum Exponential eliminated observed linkability risk on the datasets. Formal statistical testing using paired Wilcoxon signed-rank tests with Holm–Bonferroni correction across 156 cross-dataset comparisons found 133 significant results (85.3%), with large paired effect sizes for the Laplace and Anisotropic Gaussian quantum variants. These results demonstrate that structured quantum-inspired perturbations can substantially improve privacy–utility trade-offs while maintaining robustness across independent real-world mobility datasets.