A hybrid BiLSTM-transformer-GCN architecture with API fusion for adaptive transportation resource analytics

Intelligent Transportation Systems (ITS) generate large volumes of heterogeneous spatiotemporal data from connected vehicles, road infrastructure, and external contextual services. Effectively discovering, classifying, and managing these transportation resources in real time remains challenging due to dynamic traffic conditions and nonlinear dependencies. This paper proposes a novel hybrid deep learning framework that integrates Bidirectional Long Short-Term Memory (BiLSTM), Transformer encoders, and Graph Convolutional Networks (GCN) with API-based contextual feature fusion for adaptive resource discovery and classification. A multi-objective learning strategy jointly optimizes supervised classification and unsupervised anomaly discovery through reconstruction-based latent feature refinement. Experiments conducted on the Barcelona Accident dataset and the Metro Interstate Traffic Volume dataset demonstrate the superiority of the proposed method. On the combined dataset, the framework achieves an accuracy of 0.882, F1-score of 0.864, and AUC of 0.911, outperforming traditional baselines such as Random Forest (accuracy 0.839) and deep CNN-LSTM (accuracy 0.857). For anomaly discovery, it obtains an AUROC of 0.836 and AUPRC of 0.784 on the Barcelona dataset, and AUROC of 0.819 on the Metro dataset, while achieving the lowest reconstruction error of 0.198. The ablation study shows performance degradation when API fusion (-4.1% accuracy) or GCN modeling (-2.8% accuracy) is removed, confirming their contribution. Inference latency analysis further demonstrates practical deployability with a competitive accuracy-latency trade-off.