Classifying Road Accidents and Forecasting Level of Risk Based on a Combined PCA-LPP and DBSCAN Method
-
摘要: 道路交通事故是全球范围内造成大量人员伤亡和财产损失的重大问题之一，通过对道路交通事故进行分类和风险等级预测，能够锁定高风险车辆，以减小事故的发生和人员伤亡的概率。交通事故往往由环境、天气、道路条件、路段设施等多维特征相互作用形成，现有的事故影响分析方法缺乏对交通事故数据的综合研究。为此本文提出1种交通事故分类模型，在传统PCA算法的基础上通过衡量各等级数据间的相似性对数据集进行二次降维，采用改进后降维算法PCA-LPP处理大规模交通事故数据集；利用DBSCAN算法对事故数据划分风险区域，根据迭代训练出的各等级空间对模拟车辆环境进行风险划分。试验结果表明：在大规模交通数据降至不同维度的对比实验中，证明PCA-LPP算法使降维后的特征与样本的类别相关程度更高；同时，利用基于密度的DBSCAN聚类算法处理复杂且伴有偶发性的交通事故数据时，算法的纯度为0.942 9、兰德指数为0.946 2，互信息指数为0.678 4，与K-means、谱聚类等传统算法结果相比，DBSCAN算法的各项评估指标均高于其他算法，从分类效果图发现该模型减少了噪声数据的影响；最后，通过消融实验验证了带有二次降维的PCA-LPP算法的各项评估指标均为最高。其预测结果的混淆矩阵显示该模型对各风险等级的精确率分别为85.77%、70.78%、80.65%，验证了模型的有效性与实用性。Abstract: Road traffic accidents are one of the major problems causing large numbers of casualties and property losses worldwide. By classifying road traffic accidents and predicting risk levels, it becomes possible to identify high-risk vehicles and reduce the probability of accidents and casualties. Traffic accidents are often influenced by multiple factors such as environment, weather, road conditions, and infrastructure, but existing accident impact analysis methods lack comprehensive research on traffic accident data. Therefore, this paper proposes a traffic accident classification model that incorporates an improved dimensionality reduction algorithm called PCA-LPP, which measures the similarity between data of different levels to achieve secondary dimensionality reduction. The model utilizes a large-scale traffic accident dataset and applies the DBSCAN algorithm to partition the accident data into risk areas. By training the spatial representations of different risk levels iteratively, the model could assess the risk levels in simulated vehicle environments. Experimental results demonstrate the effectiveness of the proposed approach. Comparative experiments on large-scale traffic data reduced to different dimensions show that the PCA-LPP algorithm achieves higher correlation between the reduced features and sample categories compared to traditional PCA. Moreover, when handling complex and sporadic traffic accident data, the density-based DBSCAN clustering algorithm achieves a purity of 0.942 9, a Rand index of 0.946 2, and a mutual information index of 0.678 4. Comparing these results with traditional algorithms like K-means and spectral clustering, DBSCAN consistently outperforms them in various evaluation metrics. Additionally, visual analysis of the classification results indicates that the proposed model reduces the influence of noisy data. Finally, an ablation experiment confirms that the PCA-LPP algorithm with secondary dimensionality reduction achieves the highest evaluation metrics. The confusion matrix of the prediction results shows that the model achieves precision rates of 85.77%, 70.78%, and 80.65% for different risk levels, further validating its effectiveness and practicality.
-
Key words:
- traffic safety /
- accident class classification /
- risk prediction /
- PCA-LPP algorithm /
- DBSCAN algorithm /
-
图 6 密度半径参数列表与K值的关系曲线
Figure 6. Relationship curve between radius Eps parameter list and K value
图 8 带噪声的候选密度半径与各评估指标关系图
Figure 8. Relationship between candidate density radius with noise and each evaluation index
图 9 候选密度半径与各评估指标关系图
Figure 9. Relationship between candidate density radius and each evaluation index
图 10 聚类算法分类效果对比图
Figure 10. Comparison chart of classification effect of clustering algorithm
表 1 事故影响因素统计
Table 1. Statistics of accident influencing factors
类型 变量 连续变量 离散变量 最小值 最大值 均值 标准差 众数 样本量 比例/% 环境 白天 11 719 56.80 月份 1 12 6.5 4.71 12 夜晚 8 913 43.20 小时/h 0 24 11.51 5.48 7 天气 天气描述 阴天 温度/°F -8 103 35.59 11.07 41 湿度/% 9 100 74 18.65 阴天 风寒指数/°F -25.2 107 26.93 13.87 37.1 风速（/ m/s） 0 22.620 224 4.376 521 6 2.436 368 2.592 832 压力/ kpa 69.243 7 105.067 6 101.546 14 1.726 86 101.986 32 道路条件 能见度/km 0 112.654 12.713 8 5.6 9.334 2 信号灯 2 888 14 交汇口 1 485 7.2 十字路口 1 093 5.3 环岛 2 0.009 7 减速带 1 0.004 8 路障 433 0.021 路段设施 娱乐设施 256 1.24 礼让标志 536 0.26 停止牌 255 1.23 铁路 124 0.60 车站 280 1.36 
下载: 导出CSV
表 2 预处理后的数据特征
Table 2. Preprocessed data characteristics
编号 影响间距/m 温度/°F 相对湿度/% 气压/kPa 能见度/km 风速（/ m/s） … 月份 小时 0 0.01 41 81 103.002 1 16.093 44 3.084 576 … 2 3 1 1.88 41 93 99.379 1 4.023 36 5.140 96 … 4 9 2 9.48 42.1 85 101.546 14 16.093 44 1.564 64 … 5 4 $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ $\vdots$ 20 630 0.01 46.4 100 103.020 9 16.093 44 1.564 64 … 1 7 20 631 0 46.4 100 101.986 32 4.828 3.621 024 … 1 9 20 632 2.28 39.9 97 103.069 84 4.023 36 4.649 216 … 1 7
下载: 导出CSV
表 3 聚类算法性能比较
Table 3. Performance comparison of clustering algorithms
算法 Purity RI MI K-means 0.902 2 0.740 7 0.435 1 Mean Shift 0.702 1 0.768 1 0.500 7 Spectral Clustering 0.666 4 0.700 2 0.389 8 DBSCAN（含离群点）-3簇 0.833 3 0.818 2 0.625 3 DBSCAN-3簇 0.942 9 0.946 2 0.678 4 DBSCAN（含离群点）-5簇 0.856 4 0.841 1 0.665 3 DBSCAN-5簇 0.889 3 0.882 0 0.7362
下载: 导出CSV
表 4 消融实验聚类效果对比
Table 4. Comparison of clustering effects of ablation experiments
算法 密度半径 最小点数 Purity RI MI DBSCAN 720 70 0.501 9 0.511 3 0.116 8 PCA-3D+DBSCAN 348 70 0.411 7 0.520 0 0.017 5 PCA-8D+DBSCAN 704 50 0.498 4 0.499 1 0.118 6 PCA-LPP+DBSCAN 51 63 0.833 3 0.818 2 0.625 3
下载: 导出CSV
-
[1] 黎健侃, 李泽炜, 华汶雯, 等. 城市道路交通事故统计分析[J]. 科技创新与应用, 2021, 11(21): 74-76. https://www.cnki.com.cn/Article/CJFDTOTAL-CXYY202121024.htmLI J K, LI Z W, HUA W W, et al. Statistical analysis of urban traffic accidents[J]. Technology Innovation and Application, 2021, 11(21): 74-76. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-CXYY202121024.htm [2] 林宣财, 张旭丰, 王佐, 等. 基于交通事故多发位置的区间平均纵坡控制指标研究[J]. 公路交通科技, 2021, 38(9): 105-113. https://www.cnki.com.cn/Article/CJFDTOTAL-GLJK202109014.htmLIN X C, ZHANG X F, WANG Z, et al. Study on control indicator of interval average longitudinal slope based on location of traffic accidents[J]. Journal of Highway and Transportation Research and Development, 2021, 38(9): 105-113. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-GLJK202109014.htm [3] PANG J, KRATHAUS A, BENEDYK I, et al. A temporal instability analysis of environmental factors affecting accident occurrences during snow events: The random parameters hazard-based duration model with means and variances heterogeneity[J]. Analytic Methods in Accident Research, 2022, 34: 100215. doi: 10.1016/j.amar.2022.100215 [4] 陈昭明, 徐文远. 基于负二项分布的高速公路交通事故影响因素分析[J]. 交通信息与安全, 2022, 40(1): 28-35. doi: 10.3963/j.jssn.1674-4861.2022.01.004CHEN Z M, XU W Y, An analysis of factors influencing freeway crashes with a negative binomial model[J]. Journal of Transport Information and Safety, 2022, 40(1): 28-35. (in Chinese) doi: 10.3963/j.jssn.1674-4861.2022.01.004 [5] 吴仁彪, 何宇翔, 贾云飞. 基于改进层次分析法的重点人员风险评价方法[J]. 中国安全科学学报, 2021, 31(10): 112-118. https://www.cnki.com.cn/Article/CJFDTOTAL-ZAQK202110020.htmWU R B, HE Y X, JIA Y F. Key personnel risk assessment method based on improved AHP[J]. China Safety Science Journal, 2021, 31(10): 112-118. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-ZAQK202110020.htm [6] AIASH A, ROBUSTE F. Traffic accident severity analysis in Barcelona using a binary probit and CHAID tree[J]. International Journal of Injury Control and Safety Promotion, 2022, 29(2): 256-264. doi: 10.1080/17457300.2021.1998136 [7] LI W, ZHAO X, LIU S. Traffic accident prediction based on multivariable grey model[J]. Information, 2020, 11(4): 184-196. doi: 10.3390/info11040184 [8] HUSSAIN F, LI Y, ARUN A, et al. A hybrid modelling framework of machine learning and extreme value theory for crash risk estimation using traffic conflicts[J]. Analytic Methods in Accident Research, 2022, 36: 2213-6657. [9] GILANI V N M, HOSSEINIAN S M, GHASEDI M, et al. Data-driven urban traffic accident analysis and prediction using logit and machine learning-based pattern recognition models[J]. Mathematical Problems in Engineering, 2021, 2021(10): 1-11. [10] 张洁, 张萌萌, 李虹燕. 基于二元Logistic模型的城市道路交通事故严重程度分析[J]. 交通信息与安全, 2022, 40 (5): 70-79. doi: 10.3963/j.jssn.1674-4861.2022.05.008ZHANG J, ZHANG M M, LI H Y, An analysis of severity of traffic accidents on urban roadways based on binary logistic models[J]. Journal of Transport Information and Safety, 2022, 40(5): 70-79. (in Chinese) doi: 10.3963/j.jssn.1674-4861.2022.05.008 [11] MOUSAVIAN ANARAKI S A, HAERI A, MOSLEHI F. A hybrid reciprocal model of PCA and K-means with an innovative approach of considering sub-datasets for the improvement of K-means initialization and step-by-step labeling to create clusters with high interpretability[J]. Pattern Analysis and Applications, 2021, 24(3): 1387-1402. [12] 张志恒, 李超. 基于PCA-BP神经网络的审计风险识别研究[J]. 重庆理工大学学报(自然科学), 2021, 35(5): 253-261. https://www.cnki.com.cn/Article/CJFDTOTAL-CGGL202105033.htmZHANG Z H, LI C. The research of audit risk identification based on PCA and BP neural network[J]. Journal of Chongqing University of Technology(Natural Science), 2021, 35 (5): 253-261. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-CGGL202105033.htm [13] ISLAM M R, JENNY I J, NAYON M, et al. Clustering algorithms to analyze the road traffic crashes[C]. International Conference on Science & Contemporary Technologies(ICSCT), University of Da Nang, Vietnam: IEEE, 2021. [14] PEIXOTO M L M, MAIA A H O, MOTA E, et al. A traffic data clustering framework based on fog computing for VANETs[J]. Vehicular Communications, 2021, 2021(31): 100370. [15] BANDYOPADHYAY S, THAKUR S S, MANDAL J K. Product recommendation for e-commerce business by applying principal component analysis(PCA)and K-means clustering: benefit for the society[J]. Innovations in Systems and Software Engineering, 2020, 17(1): 45-52. [16] MAHA A R, LOAY E G. Stamps extraction using local adaptive K-means and ISODATA algorithms[J]. Indonesian Journal of Electrical Engineering and Computer Science, 2021, 21(1)137-145. [17] 赵莉, 候兴哲, 胡君, 等, 基于改进K-means算法的海量智能用电数据分析[J]. 电网技术, 2014, 38(10): 2715-2720. https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS201410015.htmZHAO L, HOU X Z, HU J, et al. Improved K-means algorithm based analysis on massive data of intelligent power utilization[J]. Power System Technology, 2014, 38(10): 2715-2720. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS201410015.htm [18] LU X, LONG J, WEN J, et al. Locality preserving projection with symmetric graph embedding for unsupervised dimensionality reduction[J/OL]. (2022-06)[2023-09-15]. https://doi.org/10.1016/j.patcog.2022.108844 .[19] CHEN Y, XIAO X, HUA Z, et al. Adaptive transition probability matrix learning for multiview spectral clustering[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(9): 4712-4726. [20] 李文杰, 闫世强, 蒋莹, 等. 自适应确定DBSCAN算法参数的算法研究[J]. 计算机工程与应用, 2019, 55(5): 1-7. https://www.cnki.com.cn/Article/CJFDTOTAL-JSGG201905002.htmLI W J, YAN S Q, JIANG Y, et al. Research on method of self-adaptive determination of DBSCAN algorithm parameters[J]. Computer Engineering and Applications, 2019, 55 (5): 1-7. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-JSGG201905002.htm [21] SALIH HASAN B M, ABDULAZEEZ A M. A Review of principal component analysis algorithm for dimensionality reduction[J]. Journal of Soft Computing and Data Mining, 2021, 2(1): 20-30. -
点击查看大图
图(11) / 表(4)
计量
- 文章访问数: 704
- HTML全文浏览量: 398
- PDF下载量: 47
- 被引次数: 0