A Multilayer Perceptron Neural Network with Selective-Data Training for Flight Arrival Delay Prediction

Table 1

Summary of flights delay prior studies.

Reference	Objective	Study case	Methods	Features	Results
[21]	Delay deviation time prediction	Lithuanian airports	Probabilistic neural network (PNN) MLP, decision trees (DT), random forest (RF), tree ensemble (TE), GB trees, SVM	Flight date, flight number, company, departure airport, destination airport, temperature, sky information, wind speed, wind angle, visibility, scheduled time, and classes	DT, RF, TE = 62% for arrival delays and 96.02% for departure. PNN, MLP = 96.02% on the departure dataset. MLP = 47% on arrival dataset. GB trees = for arrivals 88.59%; for departures 96.02%. SVM = 32.70% for arrivals and 82.88% for departures
[16]	A comprehensive study of traffic delays	US flights	MLP, CNN	—	MLP: 89.07%, CNN: 89.32%
[23]	Flight delay prediction	US main airports	RF classifier, delay propagation model	BTS dataset features, NOAA meteorological features, scheduled departures, arrivals, direction, airport	Arrival accuracy: 0.85 departure accuracy: 0.82
[30]	Flight departure delay prediction	Beijing Capital International Airport (PEK)	Improved SVM, KNN and RF	Delay time, airline, scheduled departure time, scheduled arrival airport, type of aircraft, departure direction, scheduled flight duration, order of flight in task ring, duration of ground service, median of flight’s historical delay time, standard deviation of the flight’s historical delay time	: improved SVM = 0.71, KNN = 0.14, RF = 0.09
[31]	Flight delay prediction for commercial air transport	Beijing Capital International Airport (PEK)	DBN-SVR, kNN, SVM, LR	Airlines, air traffic control in formation, types of aircrafts, check-in and closing time of flights, boarding and taking-off time, parking area, gate locations, runway, and fuel filling time	: DBN-SVR = 0.93, kNN = 0.87, SVM = 0.87, LR = 0.82
[18]	Prediction of air delays	The route linking São Paulo (Congonhas) to Rio de Janeiro (Santos Dumont)	ANN with random search technique	—	Accuracy superior of 90%
[20]	Flight delays prediction	Hartsfield-Jackson Atlanta International Airport	Decision trees, random forest, MLP	Flight data, weather data, airplane info, delay propagation information	MLP: 85.63% accuracy, RF: 84.26%, DT: 83.06%
[22]	A cost-sensitive delay prediction	US domestic flights	Decision trees, random forest, adaboost, and kNN	Origin, destination, quarter of year, month, day of month, day of week, scheduled departure and arrival times, arrival delay indicator, NOAA meteorological data	RF = 82.75%, ada = 83.07%. kNN = 80.6%. DT = 82.49%
[17]	Airborne delay predict due to air traffic control	Tokyo Airport	ANN, queuing analysis method	Departure time, estimated time of arrival, estimated time to enter sector, sector entrance point, forecasted wind, number of aircraft in sector, sector entrance time interval	ANN RMSE = 183.7 seconds, queuing RMSE = 230.4 seconds
[24]	Departure and arrival flight delays prediction	An individual airport case in US	Gradient boosted decision tree algorithm	Year, month, day of month, day of week, carrier, origin airportID, dest airportID, CRSDepTime, DepDelay, DepDel15, CRSArrTime, ArrDelay, ArrDel15, and Cancelled	Arrival: 92.31%. departure: 94.85%
[26]	Prediction of on time flights performance	Domestic flights of the USA	GB, RF, adaboost, extra trees and MLP classifiers and regressors	Airline ID, flight number, origin airport ID, destination airport ID, year, quarter of year, month, day of month, day of week, scheduled departure time, scheduled arrival time, wind direction, humidity, pressure, temperature	Between 85% and 94% depending on whether it is a classifier or regressor
[27]	Flight arrival delays prediction	United States, 2015	Decision tree, logistic regression neural networks classifiers	Month, day, day of the week, flight number, origin airport, destination airport, scheduled departure, departure delay, taxi-out, distance, scheduled arrival	91% accuracy
[28]	Flights delay prediction modeling	US domestic flights	Multiple linear regression, decision trees, random forest	Departure delay, taxi in, taxi-out, carrier delay, security delay, weather delay, late aircraft delay, distance, and national air system delay	RMSE for DT = 26.5 minutes, MLR = 21.2 minutes, RF = 12.5 minutes
[19]	Incoming flights delays prediction	John F. Kennedy Airport	Multilevel input layer ANN	Day of month, day of week, code of origin, scheduled departure time, actual departure time, departure delay, scheduled arrival time, actual arrival time, arrival delay, carrier delay, weather delay, NAS delay, security delay, late aircraft delay	RMSE = 0.1366
[29]	Flight departure delay analysis	A large hub airport case	Bayesian network	Flight terminal number, airlines, flight task, airplane type, international (I) or domestic (D) flights, flight departure time duration, departure delay time	Accuracy is around 84.01% and 89.5% depending on the algorithm