Research On Flight Delay Prediction And Aviation Customer Churn Based On Complex Spatiotemporal Perspectiv

The rapid growth of the economy has brought development on the air traffic demand.Compared with the high-speed growth of aviation industry scale,the airport facilities,number of airlines and routes,and airspace increase at a slow pace.The unbalanced development between the two has brought a huge pressure on the aviation system,and increased the frequency of flight delays,which in turns affecting the healthy development of civil aviation enterprises.Besides,with the growth of aviation industry scale,the customer’s complaints are also increasing,which further affects the market profits of aviation companies.Consequently,controlling and reducing flight delays and associated customer complaints are an important task for maintaining the healthy development of civil aviation enterprises.Analyzing the flight and customer data,discovering the main characteristics of flight delays and customer’s complaints,and constructing a reliable flight delay and customer churn prediction model play an important role for improving the market competitiveness of civil aviation enterprises.Based on machine learning theory,this work focuses on flight delays and customer churn and explores the impact mechanism of flight delays and customer satisfaction from "flight delay prediction" → "delay propagation analysis" and "customer satisfaction research" → "customer churn prediction" perspective,provides technical support for civil aviation companies to build intelligent flight delays and customer churn early warning systems.The core content of this work is as follows.(1)Flight delay prediction from spatial-temporal perspective.In the field of flight prediction,current research focuses on mining more potential flight delay factors by adding data dimensions to achieve prediction performance improvement.And due to the connectivity of the aviation system and the continuity of flight operations,there is an obvious spatiotemporal correlation in the flight delay problem.Therefore,this paper constructs a flight delay prediction model considering the spatio-temporal relationship from the perspective of time and space to achieve real-time monitoring of the flight delay problem.First,from the spatial and temporal perspective,we use the complex network theory and LSTM algorithm to extract the spatial and temporal features and apply Random Forest as classifier to predict flight delays.Second,due to the spatial-temporal dependency relationship of flight data,the spatial-temporal properties of flight delays cannot be understood from information at the individual spatial or temporal level alone.In this work,the convolutional neural network(CNN)and long short-term memory network(LSTM)are combined to construct a CNNLSTM framework to extract the spatiotemporal dependency model of flight data,and use random forest algorithm to predict flight delays.Third,by assessing the individual effect of weather and non-weather features on flight delays,we update the random forest algorithm to a probabilistic forest algorithm that considers the priority information of different features for flight delay prediction.Experiments are carried out on different airports and the comparison results demonstrate that the proposed model outperformance the prior research.(2)Modeling delay propagation from spatial-temporal perspective.Due to the connectivity of the aviation system,delays at a single airport can be affected by multiple airports upstream.And the existing studies mainly analyze the delay causality between origin-destination airport pairs,ignoring the interaction effect between multiple airports.Therefore,this paper takes the whole airport network as the research object,constructs the delay causality between airport networks,and analyzes the characteristics of delay propagation from two time-space dimensions.First,using the real flight data,we build a delay propagation network based on Bayesian network.Through topological analysis of the delay propagation network,we explore the characteristics of delay propagation.Second,we model the process of delay propagation from spatial and temporal perspective and identify the impact of local area delays on others,and the dynamic process of delay propagation.Third,this work investigates prevention strategies from impact effect and network structure level,and provides theoretical guidance for alleviating delay propagation in aviation networks.(3)Analysis of airline customer satisfaction.Flight delay problems can significantly reduce customer satisfaction and lead to customer churn.Therefore,in this part,we take flight delay-customer satisfaction as the research object,analyze the influence of flight delay,service quality and other factors on customer satisfaction,construct an interactive,nonlinear relationship model of flight delay,service quality and customer satisfaction,and simulate and analyze the effect of flight delay on customer satisfaction.In this section,we use the Bayesian network as basic model,and use the real customer data of airline for experiment to construct a model that describes the relationship among flight delays,service quality and customer satisfaction.Moreover,we explore the influence of different factors on airline customer satisfaction,identify the key factors that affects customer satisfaction.At last,we investigate the improvement of flight delays and service quality on customer satisfaction,and provide a basis for building airline customer management theories.(4)Airline customer churn prediction.Customer dissatisfaction experience can lead to customers choosing other airlines’ services,which in turn causes a customer churn problem and leads to some impact on the market demand of airline companies.Therefore,in this work,we propose an improved Bagging algorithm based on satisfaction and by considering the probability of customer churn to simulate the customer churn problem and analyze the impact of flight delays on customer churn.First,based on customer satisfaction research,the potential probability of customer churn is determined,and a variant of Random Forest model is proposed to predict customer churn.Second,we analyze the importance of features that affect customer churn and sort them to identify the key influencing features.Third,the established customer churn prediction model is used to simulate the impact of flight delays and service quality improvement on the number of customers churn,and provide scientific advice for companies to develop customer retention strategies.