Visual Analytics Of Traffic Congestion Influencing Factors Based On Explainable Machine Learning

In the process of urbanization,traffic congestion is becoming increasingly serious and widespread,which not only increases the travel time and cost of people,but also affects travel safety,seriously hindering the rapid development of social economy,and has become an urgent livelihood issue to be resolved.How to effectively alleviate traffic congestion is one of the most important challenges facing the urban transportation field,and accurately grasping the influencing factors of traffic congestion is the core part of alleviating traffic congestion issues.However,how to accurately detect traffic congestion conditions,quickly identify and perceive the multi-source influencing factors of traffic congestion is a huge challenge.On the one hand,the road traffic state is time-varying,and the road network structure has complex spatial dependencies,resulting in very complex identification of traffic congestion states.On the other hand,traffic congestion is affected by many factors such as road network structure,weather,and holidays.There are potential correlations among the influencing factors,making effective attribution analysis of the influencing factors extremely difficult.In addition,it is extremely important to gain insight into the spatiotemporal variation of traffic congestion and identify the key characteristics of multi-source influencing factors for quickly sensing traffic conditions and formulating reasonable traffic dispersion plans.In the context of the big data era,many scholars,from the perspective of data science,have constructed in-depth learning models to accurately predict traffic congestion,thereby providing a basis for early warning and mitigation of traffic congestion.However,most of these models lack interpretability and credibility,seriously affecting the accuracy of traffic congestion mitigation decisions.In view of this,this paper introduces a visual analytics method that integrates humanmachine intelligence,proposes a traffic congestion attribution analysis method based on interpretable machine learning,establishes a visual analytics framework for road traffic conditions from overall overview to detailed exploration,and provides transportation experts with scientific basis and practical means to alleviate road congestion.The main research content of this paper is as follows.(1)A road speed prediction model based on multi-source spatiotemporal data fusion is proposed.This paper considers the multi-source influencing factors of road speed,and uses the technology of multi-source heterogeneous information fusion to quantify the influencing factors of road speed.From the perspective of multi-source influencing factors,it proposes a road speed prediction model that integrates multi-source spatiotemporal data,perceives urban traffic conditions,extracts congested sections,and discovers the spatiotemporal changes in congestion,assisting decision makers to take timely intervention measures to alleviate traffic congestion.(2)A quantitative analysis method of road speed influencing factors based on SHAP interpretability is proposed.The SHAP method is used to model the response relationship between influencing factors and road speed,and a mapping method from multi-source influencing factors to feature factors is proposed to explore the nonlinear effects of road network structure,weather,air quality,and holidays on road speed,and to explore the internal relationship between road feature factors and the importance of their corresponding features.(3)A visual analytics framework for traffic congestion influencing factors based on interpretable machine learning is proposed.The framework can provide insight into traffic congestion through comprehensive overview,detailed exploration,multi view linkage,and rich interactive means.Not only does it support comparative analysis of factors affecting road speed across multiple roads,but it can also conduct case level exploration on a single road.This paper also designed a time series view and a spatial distribution view to discover the spatiotemporal changes in road congestion,designed and implemented a multi-level visual analytics system for traffic congestion impact factors to help decision makers explore and perceive the potential impact factors of congestion.This article has conducted five case studies using real data sets,verifying the effectiveness of the methods and systems proposed in this article in analyzing the impact factors of traffic congestion,and providing important evidence for urban traffic managers to formulate targeted traffic dispersion plans.