Flight Delay Prediction Based on Characteristics of Aviation Network

Tan Zhou,Xin Chen,Zongwei Xun,Qiang Gao,R Montemanni

doi:10.1051/matecconf/201925902006

Abstract

In recent years, the increasingly serious flight delay affects the development of the civil aviation. It is meaningful to establish an effective model for predicating delay to help airlines take responsive measures. In this study, we collect three years’ operation data of a domestic airline company. To analyse the temporal pattern of the Aviation Network (AN), we obtain a time series of topological statistics through sliding the temporal AN with an hourly time window. In addition, we use K-means clustering algorithm to analyse the busy level of airports, which makes the airport property value more precise. Finally, we add delay property and use CHAID decision tree algorithm to train the data of an airline for nearly 3 years and use the train?ing model to predicate recent half a year delay. The experimental results show that the accuracy of the model is close to 80%.

Highlights

With the developing of civil aviation, the number of airports and flights are increasing sharply
Based on the topological characteristics analyzed in Chapter 2: 1. Shorter average path length and higher clustering coefficient can both lead to higher delays, because flight delay of any node will be quickly propagated to other nodes
We add in some social characteristics of the airports and build a clustering index system

Summary

Introduction

With the developing of civil aviation, the number of airports and flights are increasing sharply. In this paper, combined with the dynamic topology characteristics of aviation network, a prediction model based on CHAID decision tree is proposed. The average shortest path length =2.412, the average clustering coefficient =0.663, and the network diameter is 6, indicating that the separation between the two airports is very small, and the average can be reached in as few as 2 transfers. It needs no more than four connections for any two farthest airport to reach each other.

Dynamic Characteristic of Aviation Network

Conclusion

Attribute selection and division

Experimental results

Comprehensive analysis