A Study On Risk Prediction Of Distracted Driving Behavior Based On Hybrid Neural Network

Distracted driving behavior is one of the key factors contributing to traffic accidents.Timely and accurate prediction of distracted driving behavior risks is beneficial for perceiving real-time traffic risks,which is an important yet challenging task in modern traffic safety management.The potential causes of accidents resulting from distracted driving pose harm risks to road traffic participants,especially in public transportation and hazardous goods transportation.With the rapid advancement of Internet of Things(Io T)technology,a large amount of driving behavior and characteristic data collected by sensors installed both inside and outside these vehicles can be obtained through algorithm design and recognition,enabling the acquisition of distracted driving behavior information.From a technical perspective,due to the spatiotemporal heterogeneity of distracted driving behavior data,it is necessary to consider both temporal and spatial dependencies when constructing prediction models.Additionally,distracted driving behavior is also influenced by external factors such as time and road conditions.Therefore,this study proposes individualbased and road network-based prediction models for distracted driving behavior risks using neural network technology.The primary research focus of this study is outlined below.(1)Design and implementation of algorithms for predicting distracted driving behavior risks in individual drivers: A model called the Driving Behavior Risk Prediction Neural Network(DBRPNN)is developed for predicting distracted driving behavior data in individuals.The research predicts driving behavior risks for different entities(vehicles and roads).To enhance the applicability of the model,the research further classifies distracted driving behavior,and the DBRPNN provides more accurate risk predictions.Results indicate that compared to traditional models such as classification and regression trees,support vector machines,recurrent neural networks,and long short-term memory neural networks,the DBRPNN demonstrates superior predictive performance.(2)Algorithm for predicting distracted driving behavior risks in road networks: Due to the complex topology and dynamic temporal changes in road networks,this research proposes a neural network-based approach called Distracted Driving Risk Prediction(DDRP)that combines deep learning and spatiotemporal dependencies.This approach accurately predicts the scale of distracted driving behavior on road networks.Experimental results demonstrate that the proposed method performs relatively well compared to other methods used in this research,such as extreme gradient boosting,long short-term memory neural networks,and a combination of convolutional neural networks and long short-term memory networks.Furthermore,this method accurately predicts the scale of distracted driving behavior for different categories,time intervals,and grid units.In summary,this article proposes individual and road network distracted driving behavior risk prediction algorithms that effectively extract spatiotemporal dependent features based on distracted driving behavior sequence data and integrate external attribute data into the features to achieve high-precision prediction results.The combination of the two can achieve a more comprehensive and accurate prediction of distracted driving behavior risk.The experiments are conducted on a large,real provincial dataset and compared with various classical and advanced models,using multiple evaluation metrics from multiple perspectives to validate the effectiveness and flexibility of the proposed models.The results provide support and reference for the analysis and understanding of traffic safety risk situations,as well as the improvement of traffic management and control decision-making means.