Automatic Clustering for Unsupervised Risk Diagnosis of Vehicle Driving for Smart Road

Early risk diagnosis and driving anomaly detection from vehicle stream are of great benefits in a range advanced solutions towards Smart Road crash prevention, although there intrinsic challenges, especially lack ground truth, definition multiple exposures. This study proposes domain-specific automatic clustering (termed AutoCluster) to self-learn the optimal models for unsupervised assessment, which integrates key steps into an auto-optimisable pipeline, including feature algorithm selection, hyperparameter auto-tuning. Firstly, based on surrogate conflict measures, series indicator features constructed represent temporal-spatial kinematical Then, we develop selection method identify useful by elimination-based model reliance importance (EMRI). Secondly, propose balanced Silhouette Index (bSI) evaluate internal quality imbalanced clustering. A loss function is designed that considers performance terms quality, inter-cluster variation, stability. Thirdly, Bayesian optimisation, auto-selection auto-tuning self-learned generate best results. Herein, NGSIM trajectory data used test-bedding. Findings show AutoCluster reliable promising diagnose distinct levels inherent generalised behaviour. We also delve clustering, such as, algorithms heterogeneity, analysis, hierarchical flows, etc. Meanwhile, labelling threshold calibration. Furthermore, tackle challenges without truth or priori knowledge.