VC-RFB: feature system matrix enhanced virtual clustering random forest-Bayesian model for real-time accident prediction

Real time traffic accident prediction is the core service of intelligent transportation. However, due to the diversity of data types, existing researches based on mathematical statistics face the problem of “curse of dimension”, which can easily lead to a decrease in model real-time performance. A feature system matrix enhanced virtual cluster Random Forest-Bayesian model is proposed to address this issue. The method utilizes feature system matrix enhanced virtual clustering and random Bayesian modules to achieve dimension reduction and region partitioning of high-dimensional data. On the PeMS and TxDOT datasets, experimental results show that VCRFB improves prediction accuracy of the model by 3% while ensuring the real-time performance compared to RFB and STIML-logit.