ISDSS—Intelligent Service Decision Support System Driven by Multimodal Big Data

This study proposes an Intelligent Service Decision Support System (ISDSS) driven by multimodal big data to address urban public service challenges in rapidly urbanizing environments. The system integrates heterogeneous data from transportation, healthcare, and environmental domains through a three-layer architecture comprising multi-source data input for aggregating structured and unstructured data, data fusion and feature encoding combining convolutional neural networks for spatial dependencies, long short-term memory networks for temporal dynamics, and cross-modal attention mechanisms for dynamic feature weighting, and prediction and optimization layers generating demand forecasts and resource allocation strategies. Key innovations include a multimodal fusion framework resolving data heterogeneity and a hybrid model integrating CNN, LSTM, and attention mechanisms to enhance interpretability. Empirical validation on datasets from Los Angeles and New York demonstrates superior performance: the proposed model reduces MAE by 29.1% compared to LSTM and 18.1% against GraphWaveNet in 1-h traffic prediction. Ablation experiments confirm the critical roles of attention mechanisms and external data integration, with their removal increasing MAE by 12.9% and 9.1%, respectively. Stability analysis reveals only 65.7% MAE growth from 15-min to 1-h predictions, outperforming benchmarks. Limitations in real-time processing efficiency and adaptability to extreme scenarios highlight future research directions.