Analysis of Aircraft Delay With Deep Belief Network and Causal Graph Methodology

Based on the causal estimation of Bayesian network and supervised DBN classification, this paper analyzes 120 million flight records in the United States from 1987 to 2008 to obtain a set of factors with high causal relationship with aircraft delay. According to the descriptive statistics of the correlation and frequency among the variables, this paper selects 6 variables to participate in the causal estimation, and finally obtains Distance, ActualElapsedTime and TaxiOut, as three factors with the greatest impact on aircraft delay. Then we take these three variables as the input of the DBN classifier, to predict the actual delay types of aircraft. The results of the DBN classification have passed the test and are very convincing. Finally, we get the three factors that have the highest causal relationship with aircraft delay. By comparing DBN classification with naive Bayesian classifier, it is found that the classification accuracy trained by DBN is higher. This paper screens the variables by the results of statistical analysis, and classifies the variables that may affect the delay of airplanes. From the data, the final results are more practical. In this paper, data are analyzed and trained by means of sampling year by year, and the results of analysis may change. In the subsequent analysis, we can use tools that can handle more data to experiment, so that the analysis can be more accurate.