Analysis Of Influencing Factors And Risk Prediction Of Aviation Safety Accident Report Based On Text Mining

Nowadays,more and more people choose to travel by plane in the world.How to improve aviation safety is the first topic faced by the major airline company in the world.The aviation safety incident report(ASRS),first established by the United States in 1970,plays an important role in analyzing the factors affecting aviation safety and provides many effective suggestions for aviation safety.The paper is based on the data of American aviation safety incidents,using text mining method to analyze the report of aviation safety incidents.Firstly,the factors affecting aviation safety are analyzed.Secondly,the risk prediction of aviation safety accident report is carried out.The process of analyzing factors affecting aviation safety includes text preprocessing,extracting original feature items,dimensionality reduction of original feature items,and building space vector model.The main steps of text preprocessing are text lowercase,punctuation removal,stop words removal and so on.The algorithm used to extract text features is TF-IDF,LDA topic model;Text original feature item dimension reduction using χ2 statistical dimension reduction method.After the above natural language processing process,26 items affecting aviation safety are finally obtained.Risk factors include six main factors and 20 general factors.The main factors are fatigue driving,misoperation,airport,aircraft,weather,environment,etc.The common factors are staffing,false command,company policy,communication,collision risk,manuals,etc.The second part of this article is based primarily on the use of traditional machine learning and several different classifiers based on Deep learning for risk prediction.The experimental results for traditional machine learning are compared with those for Deep learning.The traditional classifiers used are Random Forest,Naive Bayesian and Support Vector Machine.The classifier models based on Deep learning are Convolutional Neural Network and Circular Convolutional Neural Network.In the experiment part of Convolution Neural Network,the optimal parameters are set by the control variable method.In the process,ten times experiments are done by the tenfold cross validation method,and the average values of ten experiments are taken as the results.The main parameters to evaluate the performance of various classifiers are accuracy,recall and F1.The results show that the classification model based on Deeplearning is better than the traditional machine learning classification model.The innovation of this paper is to analyze the aviation safety accident report by using the method of deep learning.Experiments show that the neural network classification model based on deep learning has low dependence on the selection of artificial features and can automatically extract text features.Its overall classification effect is better than that of traditional machine learning and optimizes the classification effect.