Sparse Spatio-Temporal Dynamic Hypergraph Learning for Traffic Accident Prediction

Abstract

Traffic accidents have become one of the biggest public health safety matters, which has raised many concerns from citizens and city managers. Accurate traffic accident prediction can not only assist the government in making decisions in advance but also enhance public trust in public safety. Conventional spatio-temporal prediction models, limited by the skewed distributions and sparse labels of traffic accident occurrence, are prone to overfitting. Inspired by hypergraph learning and self-supervised learning, this paper suggests a sparse spatio-temporal dynamic hypergraph learning (SST-DHL) framework to capture the higher-order dependencies in sparse traffic accidents. Specifically, a multi-view spatio-temporal convolution block is employed first to capture the local spatio-temporal correlation and inherent semantics of traffic accidents. Then we propose a cross-regional dynamic hypergraph learning model to capture global spatio-temporal dependencies beneath the entire urban landscape. In addition, a two-supervised self-learning paradigm is intended to strengthen the representation of sparse traffic occurrences by regional self-identification, which can capture local and global spatio-temporal traffic patterns. The proposed model is applicable to most sparse datasets for traffic forecasts. Extensive experiments was conducted on two heterogeneous accident datasets from New York City and London, and the results shows an average improvements of 7.21%-23.09% at different sparsity levels compared to the optimal baselines. More importantly, the proposed SST-DHL improves the interpretability of model results, which demonstrates that hypergraph learning can efficiently capture the complex higher-order spatio-temporal dependencies among different traffic accident instances.

Prediction of chlorine residual in effluent based on SSA-SFA-CFBLS for water treatment process parameters

“…Finding the optimal hyperparameters for the model in the training and prediction process often requires a significant amount of resources. By using the SSA algorithm and setting the three hyperparameters as positive integers within the range of [1,20], [1,20], and [10,200], the SSA-SFA-CFBLS model can quickly and accurately determine the optimal hyperparameters, which greatly reduces the burden of training the model for researchers. Moreover, the SSA-SFA-CFBLS model has the characteristic of online learning.…”

“…The SSA-SFA-CFBLS parameters are chosen as follows. The population size is 50, the proportion of explorers is 20%, the maximum number of iterations is 5, and the range of CFBLS hyperparameters (number of feature groups, number of enhancement groups, number of nodes in each group) is [1,20], [1,20], [10,200], respectively.…”

“…At the same time, the broad learning system has the characteristics of incremental learning. These characteristics make broad learning system widely used in image recognition, biomedical engineering, traffic prediction, control and natural language processing [19,20].…”