AI-enhanced predictive analytics for assessing infrastructure resilience and urban living conditions under climate change scenarios

An AI-enhanced predictive analytics framework for assessing urban living conditions and infrastructure resilience under changing climate-change scenarios is presented in this study. The framework combines a Deep Neural Network (DNN) for high-dimensional feature extraction, a dynamically reconfigurable MycelialNet architecture to improve adaptive learning under complex environmental variability, and Climate Change CycleGAN (CC-CycleGAN) to augment and correct climatic datasets. To identify patterns of stress brought on by climate change, the Air Quality Index (AQI), ecological vulnerability indicators, and multi-city environmental datasets were standardized and spatially examined. To support more reliable predictive modelling, CC-CycleGAN was evaluated on held-out validation subsets of the AQI-augmented climate dataset. Spatial fidelity improved by 18.4% (measured as Structural Similarity Index, SSIM, between augmented and reference spatial precipitation maps), precipitation-prediction bias decreased by 21.6% (mean absolute error reduction on monthly precipitation sequences), and data diversity increased by 32% (measured as the ratio of unique feature-space samples before and after augmentation using k-nearest-neighbour coverage). With 94.8% accuracy, 93.6% precision, 94.1% recall, 94.3% F1-score, and an AUC of 0.962, the integrated DNN–MycelialNet model outperformed baseline machine-learning models by 12–17% and reduced overfitting by 23%. The findings show that reconfigurable neural architectures and AI-driven generative translation greatly increase the accuracy of climate-risk prediction, facilitating better mitigation planning and bolstering long-term climate resilience in quickly changing urban environments. Specifically, the framework predicts three categories of climate-induced urban stress: (1) High-Risk (AQI > 200, severe ecological vulnerability score > 0.75), (2) Moderate-Risk (AQI 100–200, ecological vulnerability 0.40–0.75), and (3) Low-Risk (AQI < 100, ecological vulnerability < 0.40). Spatial prediction maps generated for 15 Chinese prefectural cities indicate that 4 cities fall in High-Risk, 7 in Moderate-Risk, and 4 in Low-Risk categories under projected 2035 climate scenarios. The study’s primary conceptual contribution is a novel AI-enhanced predictive framework combining three complementary architectures — CC-CycleGAN, DNN, and MycelialNet — for the first time in integrated urban climate-risk assessment. This framework is distinct from prior single-model or dual-model climate studies in that it addresses data augmentation, deep feature learning, and adaptive classification simultaneously within one end-to-end pipeline. Not applicable.