重庆理工大学学报2024，Vol.38Issue(15)：197-207,11.DOI:10.3969/j.issn.1674-8425(z).2024.08.023
采用VAE-CatBoost的高速公路交通事件检测框架
A framework for highway traffic incident detection based on VAE-CatBoost
摘要
Abstract
To address the feature scarcity and sample imbalance in traffic incident detection,this paper proposes a traffic incident detection framework based on variational auto-encoder-random forest-categorical gradient boosting tree is proposed.First,a more comprehensive initial feature set is built based on four rules.Second,a variational auto-encoder is employed to balance the datasets.Then,the Random Forest algorithm is employed to filter the best input feature set.Finally,the categorical gradient boosting tree algorithm is introduced as a classifier to detect traffic incidents.Experiments are conducted by using real-world traffic datasets and six effective evaluation metrics are selected to evaluate the experimental data.Our results show the proposed framework delivers the optimal performances in all evaluation metrics except the false alarm rate,indicating its superiority in traffic incident detection.关键词
交通事件检测/特征扩展/数据平衡/特征选择/梯度提升树Key words
traffic incident detection/feature expansion/data balancing/feature selection/gradient boosting tree分类
交通运输引用本文复制引用
张兵,邹少权,陆春霖,陈渤文,薛运强..采用VAE-CatBoost的高速公路交通事件检测框架[J].重庆理工大学学报,2024,38(15):197-207,11.基金项目
国家自然科学基金地区基金项目(52162042) （52162042）
江西省省教育厅一般课题(GJJ190331) （GJJ190331）
