Short-term Traffic Flow Prediction Based On Improved XGBoost

With the rapid development of the economic level,the demand for road traffic has also increased.However,urban traffic congestion has led to a reduction in vehicle speed,longer travel time,increased exhaust emissions,and increased travel costs,which directly become an important issue restricting urban development.one.Therefore,the research on short-term traffic flow prediction is of great significance to the construction and development of urban intelligent transportation systems.Short-term traffic flow prediction is the process of predicting the traffic flow at the next time interval based on historical traffic flow data.It not only reduces traffic congestion and improves transportation efficiency,but also provides an effective basis for urban transportation planning.With the development of artificial intelligence,more machine learning models and deep learning models have begun to be applied to traffic flow prediction.The machine learning algorithm makes the regression problem converge faster,and the fitted model can achieve higher accuracy.In the previous traffic flow prediction,the models used are all based on relatively shallow structure algorithms,but the models built by the weak learners do not reflect well in reality.Later,integrated algorithms appeared and were well known to everyone.It is boosting algorithm,bagging algorithm and random forest algorithm.The integrated algorithm is to combine each weak learner according to some optimal strategy to form a strong learner,this strong learner can better fit the model.In this paper,we choose the extreme gradient boosting(XGBoost)algorithm for traffic flow prediction.The XGBoost algorithm has a fast convergence speed,and the convergence speed is fast without losing the robustness of the model.In previous studies,the Mean Squared Error(MSE)is generally the default objective function of the XGBoost algorithm.In fact,the choice of objective function is very important in the application of XGBoost.For different types of data,different objective functions should be selected to build the model.In this paper,Huber loss function is selected as our objective function.Through the analysis of the periodicity and correlation of traffic flow data,the six feature attributes of monthly cycle feature,weekly cycle feature,first 3 days feature,first 2 days feature,first 1 day feature,and first5 minute feature are selected and their Perform feature importance analysis.Using this model and the extracted features for traffic flow prediction,the experimental results show that the model is superior to the extreme gradient ascent model based on mean square error loss and the extreme gradient ascent model based on average absolute error loss.At the same time,the model has higher prediction accuracy than gradient-lift regression model and support vector machine regression model,each error index is small,and the model training time is short,which meets the timeliness required by short-term traffic flow prediction.