Research On Flight Delays Prediction Methods Based On Machine Learning

In recent years,with the rapid development of the air transport industry,the number of flights has increased rapidly.However,the frequent occurrence of delays has severely disrupted the normal operation of flights.The occurrence of delays involves many aspects of flight operations.In order to solve this problem,it is necessary to dig out these hidden causal relationships and choose appropriate methods to effectively predict delays.Therefore,this article takes flight delays as the research object,uses machine learning methods to build a model,and combines relevant data to conduct experimental analysis to verify the feasibility of the prediction model,so as to provide reference for decision-making by related departments.The main research contents of this paper are as follows:First,it analyzes the hazards of flight delays to civil aviation operations,and clarifies the research content and ideas based on the current status of domestic and foreign research.Then,starting from the basic flight operation process,the influencing factors at each stage are explored and analyzed from multiple angles.Based on the specific causes of the delay,the experimental design ideas of this article are put forward;then the original data of flight operation delay data and weather monitoring data are extracted from the databases of the two major organizations of the United States BTS and NOAA,and then cleaned,coded and fused,The flight delay prediction experimental data set was obtained;finally,the prediction model was constructed by the two machine learning methods of CatBoost decision tree and LSTM long short-term memory network,and the two data sets before and after the fusion of weather information were input into the model for training.And choose three indicators of prediction accuracy,loss value and confusion matrix to analyze the quality of the experimental simulation results.Experimental results show that after adding weather information,the final CatBoost prediction accuracy reaches 84.50%,while the LSTM prediction accuracy reaches92.82%,and the fitting effect is ideal.The main innovations of this article can be summarized as follows:(1)The experiment uses more dimensions of flight operation and weather data characteristics,which can make the model learn more fully,and can deeply explore the potential connections between various influencing factors and flight delays during the training process.(2)Compared with the previous prediction methods,the machine learning method used in the experiment has a great improvement.Among them,CatBoost,as an improved decision tree algorithm,effectively improves the training speed,and the parameter adjustment process is relatively simple;compared with the traditional loop The addition of neural network and LSTM structure layer enables the model to effectively filter information and improve the accuracy of model classification.