Flight Dataset Research Articles

Quantum-Accelerated Flight Selection: Probing Grover's Algorithm and Quantum Device Efficiency

Current flight search platforms primarily consider four essential factors when planning a trip: departure/arrival dates, as well as the origin and destination locations. However, when additional parameters are added to this search, the problem shifts from a simple to a complex search, as the engine must sift through a massive dataset of flights, including information on airlines, flight routes, fees, and more. To address this challenge and improve flight search efficiency amidst data-intensive and resource-demanding environments, this paper proposes the use of Grover's search algorithm. This algorithm is demonstrated as the optimal solution for searches with increased constraints. The paper highlights the practical application of Grover's search algorithm across three datasets of assorted sizes, as well as various quantum hardware and simulators available in the NISQ era. Furthermore, this paper provides an in-depth understanding of the complexity of quantum circuit design, including the key phases of state encoding and amplitude amplification. The effectiveness of these approaches is evaluated through analysis of execution times and quantum measurement results. The aim is to showcase the potential of quantum computing in revolutionizing real-world search tasks, particularly in the realm of flight selection.

Experimental backward integration for state-dependent differential Riccati equation (SDDRE): A case study on flapping-wing flying robot

Backward integration (BI) is a classical approach for solving optimal control arising from linear quadratic regulator (LQR) design in differential form. It proposes a two-round solution, the first one starting from a final boundary condition to the initial one to generate the optimal gain, and the next one, solving the control system in a forward loop. Implementation of the BI on the nonlinear optimal control, the state-dependent differential Riccati equation (SDDRE), is a challenge since in the backward loop, the state information is missing and is not straightforward like the LQR. Hence, a control for backward motion is required to regulate the system from the terminal to the initial desired states. While there have been some valuable works on the theoretical implementation of the BI in simulations for the SDDRE, this approach has not been reported in experiments for the best knowledge of the authors. Here in this work, the SDDRE is solved using the BI and two backward and forward solution steps, and it is experimentally applied for a flapping-wing flying robot (FWFR). The execution of the control system is done onboard using Raspberry Pi 4B as the processor which has limited computational capacity. The trajectory tracking of a line in a closed limited space is proposed for the FWFR flight. The objective is to position the robot bird at the corner of the rectangular limited space with minimum error in translation. The results have been presented for 21 flights to show the repeatability and presented the best-case minimum error of 10 (cm) at the end of the trajectory, in the YZ-plane. Considering approximately 12 (m) flight path, the error is found less than 1 percent of the travel distance. The results were compared with forward integration to confirm the correctness of the computation. The experimental flight dataset, MATLAB simulation codes, and experimental Python codes are available as supplementary material for this work in the online version of the paper.

SLKIR: A framework for extracting key information from air traffic control instructions Using small sample learning

In air traffic control (ATC), Key Information Recognition (KIR) of ATC instructions plays a pivotal role in automation. The field's specialized nature has led to a scarcity of related research and a gap with the industry's cutting-edge developments. Addressing this, an innovative end-to-end deep learning framework, Small Sample Learning for Key Information Recognition (SLKIR), is introduced for enhancing KIR in ATC instructions. SLKIR incorporates a novel Multi-Head Local Lexical Association Attention (MHLA) mechanism, specifically designed to enhance accuracy in identifying boundary words of key information by capturing their latent representations. Furthermore, the framework includes a task focused on prompt, aiming to bolster the semantic comprehension of ATC instructions within the core network. To overcome the challenges posed by category imbalance in boundary word and prompt discrimination tasks, tailored loss function optimization strategies are implemented, effectively expediting the learning process and boosting recognition accuracy. The framework's efficacy and adaptability are demonstrated through experiments on two distinct ATC instruction datasets. Notably, SLKIR outperforms the leading baseline model, W2NER, achieving a 3.65% increase in F1 score on the commercial flight dataset and a 12.8% increase on the training flight dataset. This study is the first of its kind to apply small-sample learning in KIR for ATC and the source code of SLKIR will be available at: https://github.com/PANPANKK/ATC_KIR.

Data-driven multivariate regression-based anomaly detection and recovery of unmanned aerial vehicle flight data

Abstract Flight data anomaly detection is crucial to ensuring the safe operation of unmanned aerial vehicles (UAVs) and has been extensively studied. However, the accurate modeling and analysis of flight data is challenging due to the influence of random noise. Meanwhile, existing methods are often inadequate in parameter selection and feature extraction when dealing with large-scale and high-dimensional flight data. This paper proposes a data-driven multivariate regression-based framework considering spatio-temporal correlation for UAV flight data anomaly detection and recovery, which integrates the techniques of correlation analysis (CA), one-dimensional convolutional neural network and long short-term memory (1D CNN-LSTM), and error filtering (EF), named CA-1DCL-EF. Specifically, correlation analysis is first performed on original UAV flight data to select parameters with correlation to reduce the model input and avoid the negative impact of irrelevant parameters on the model. Next, a regression model based on 1D CNN-LSTM is designed to fully extract the spatio-temporal features of UAV flight data and realize parameter mapping. Then, to overcome the effect of random noise, a filtering technique is introduced to smooth the errors to improve the anomaly detection performance. Finally, two common anomaly types are injected into real UAV flight datasets to verify the effectiveness of the proposed method.

GPS-free roll and pitch estimation through pressure altitude aided inertial navigation system for a jet aircraft

This paper presents a novel approach for estimating roll and pitch angles of an aircraft by regulating the dead-reckoning drift using pressure altitude measurements. The proposed algorithm uses measurements from rate gyroscopes, pressure altitude, and an air data system to estimate the roll and pitch angles of an aircraft. This unique combination of sensor data offers a new alternative to traditional navigation approaches relying on accelerometers and satellite systems to estimate the aircraft attitude. Multiple flight data sets from a high-speed crewed US Navy T-38 jet are used to validate the proposed approach. Comparable levels of attitude estimation accuracy are reported thus demonstrating that the proposed algorithm is a viable alternative for attitude estimation purposes. Therefore, this approach can be used to improve the robustness and accuracy of attitude estimation algorithms in scenarios where GPS coverage is denied, unreliable, spoofed and/or jammed.

A novel intelligent approach for flight delay prediction

Flight delay prediction is one of the most significant components of intelligent aviation systems that may spread throughout the whole aviation network and cause multi-billion-dollar losses faced by airlines and airports, it is quickly becoming an important research issue to improve airport and airline performance. Thus this paper proposed an effective algorithm called Flight Delay Path Previous-based Machine Learning (FDPP-ML) capable of improved prediction of individual flight delay minutes using regression models to an up level of accuracy. As aviation system connectivity presents complex spatial–temporal correlations, machine learning approaches have addressed flight delay prediction by using complex flight or weather features, or private information for specific airports and airlines that are hard to obtain, In contrast, the proposed FDPP-ML improved prediction based only on basic flight schedule features even with wide flight networks. The FDPP-ML consists of a novel algorithm with a supervised learning model, which works on reshaping datasets and creates two new features the main feature is previous flight delay (PFD) for flight paths, there is a strong relationship between departure and arrival delay, and vice versa for the same flight path, which increases the strength of the training model based on historical data. For target future flights, the algorithm works on inheriting the predicted flight delay to the next flight on the same flight path and repeats this process to end the prediction forecast horizon. The proving of approach effectiveness by using a wide network of US flight arrival and departure flights containing 366 airports and 10 airlines with various metrics accuracies of regression, and explanatory the impacts on various forecast horizons 2, 6, and 12 h for future flights. The FDPP-ML outperforms traditional training models by using machine and deep learning models and improving model accuracy in 10 models with an average of up to 39% in MAE, and 42% in MSE in a forecast horizon of 2 h. Finally, providing airport and airline analysis further reveals that can improve prediction than traditional training models for the individual busiest airports "Core 30" with an average of 35% in MAE and 42% in MSE respectively, and for the busiest 10 airlines with an average of 36% in MAE and 47% in MSE respectively. The findings of this study may offer informative recommendations to airport regulators and aviation authorities for developing successful air traffic control systems for enhanced flight delay prediction to flight operational effectiveness, not only over the US flight network but with wide worldwide flight networks if a dataset of flights existed.

Performance Model Identification of the General Electric CF34-8C5B1 Turbofan Using Neural Networks

This paper presents a methodology developed at the Laboratory of Applied Research in Active Controls, Avionics, and Aeroservoelasticity to identify a performance model of the CF34-8C5B1 turbofan engine powering the CRJ-700 regional jet aircraft from simulated flight data using artificial neural networks (ANNs). For this purpose, a qualified virtual research simulator was used to conduct different types of flight tests and to collect engine data under a wide range of operating conditions. The collected data were then used to create a comprehensive database for the training of the ANN model. This process was performed using the Bayesian regularization algorithm available in the MATLAB Neural Networks Toolbox, followed by a study to identify the optimal network architecture, namely, the number of layers and the number of neurons. The validation of the methodology was accomplished by comparing the model predictions with a set of flight data collected with the flight simulator for different flight conditions and flight regimes including takeoff, climb, cruise, and descent. The results showed that the model was able to predict the engine performance in terms of fan speed, core speed, inlet turbine temperature, net thrust, and fuel flow with less than 5% relative error.

Estimation of Longitudinal and Lateral Aerodynamic Derivatives From Flight Data Using Maximum Likelihood Method

The paper presents the estimation of longitudinal and lateral (lateral-directional) aerodynamic derivatives from real flight data of Hansa-3 aircraft using Maximum Likelihood (ML) method. The application of the ML method to the flight data of an air vehicle requires the postulation of correct flight dynamic formulations having accurate description of aerodynamic model. The application of the ML method to the flight data with either the measurement noise or the process noise has been accepted as a standard approach for the estimation of the aircraft parameters. A large number of longitudinal and lateral flight data sets were gathered during the flight testing of Hansa-3 aircraft through the execution of controlled maneuvers using elevator and aileron/rudder control inputs respectively. The acquired flight data in compatible form generated for various control input forms has been analyzed during the estimation of aerodynamic derivatives using Maximum Likelihood method. The results obtained in terms of aerodynamic derivatives are reasonably accurate and have been presented in tabular and graphical form. The paper highlights the various factors affecting the estimation.

Flight departure delay forecasting

The most accurate methods to predict delays have become essential to mitigate their generation and proliferation (and prevent their spread to other airports in the network). Given that both the delay time dependence function and the input variables on which it depends are not fully known, the objective of this research is to predict the delays in the departure of scheduled commercial flights through a methodology that uses predictive tools based on machine learning/deep learning (ML/DL), with supervised training in regression, based on the available flight datasets. The novel contribution of this work is, first, to make a comparison of the predictions in terms of means and statistical variance of the different ML/DL models implemented (ten in total) and, secondly, to determine the coefficients of the importance of the features or flight attributes. Using ML methods known as permutation importance, it is possible to rank the importance of flight attributes by their influence in determining the delay time and reduce the problem of selecting the most important flight attributes. The data for the analysis was obtained from the Colombian airport system for the year 2018. From the results obtained, it is worth mentioning that the model that presents the best performance is the Ensemble, or combination method of Random Forest Regressor models, configured with 2,000 trees within the forest, with an acceptable prediction range measured with the root-mean-square-error (RMSE) metric of 16 to 33 minutes (prediction of time of the flight departure delay) depending on the scenario analysed.

An adaptive law based online system identification of cropped delta unmanned aerial vehicles from flight tests

Online system identification of an aircraft has multifaceted advantages, such as real-time health monitoring, fault detection and reconfiguration of control, to name a few. Researchers have utilized variants of the equation error method, which involves non-gradient-based computation to carry out online system identification, due to its feasibility for real-time implementation. However, it was found to be underperforming while handling high angles of attack as well as time-varying bias parameters. In the current manuscript, an attempt has been made to alleviate the aforementioned limitation by proposing a novel online system identification technique based on adaptive law and its applicability is demonstrated by using compatible flight data sets, acquired from two cropped delta wing unmanned aerial vehicles, pertaining to linear, moderately nonlinear, and near stall flight regimes. It is observed that the proposed method is able to predict unknown aerodynamic derivatives, irrespective of flight envelopes, if dynamic modes are adequately excited. Confidence in the estimated aerodynamic parameters is assured with the two-sigma error bound. It is noticed that the maximum relative standard deviation in the estimates obtained with the proposed method is less than the recursive least squares method and higher than the sequential least squares-recursive Fourier transform method in the linear flight regime. Simulated responses using the estimates obtained from aforementioned methods are compared with measured flight data, and it is witnessed that the simulated nondimensional lift coefficient with the proposed method has a relative offset of less than 34.2% w.r.t measured value in near-stall flight regimes, which is least among all three methods.

Structure, Resilience and Evolution of the European Air Route Network From 2015 to 2018

In spite of the dynamic nature of air transport, air route networks, i.e. the backbone used to organise aircraft flows, are expected to be mostly static, with small changes occasionally being introduced to improve the efficiency and resilience of the system. By leveraging a large data set of European flights comprising years 2015 to 2018, we analyse its structure and evolution from the perspective of complex networks, with the aim of firstly describing it, and secondly to confirm its static nature. Results depict a highly dynamic system, with major topological changes happening at the end of 2017. Peripheral links are usually more vulnerable, due to the lack of effective reroutings, as well as central regions; additionally, the overall resilience of the network is almost constant throughout time, in spite of an increase in traffic. We further test several hypotheses regarding the design considerations driving such evolution. Beyond specific operational insights, these results highlight the importance of taking into account the evolution of this network in the study of traffic flows.

Prediction of departure flight delays through the use of predictive tools based on machine learning/deep learning algorithms

Abstract The objective of this research is to predict the delays in the departure of scheduled commercial flights through a methodology that uses predictive tools based on machine learning/deep learning (ML/DL), with supervised training in regression, based on the available flight datasets. Since the novel contribution of this work is, first, to make the comparison of the predictions in terms of means and statistical variance of the different ML/DL models implemented and, second, to determine the coefficients of the importance of the features or flight attributes, using ML methods known as permutation importance, it is possible to rank the importance of flight attributes by their influence in determining the delay time and reduce the problem of selecting the most important flight attributes. From the results obtained, it is worth mentioning that the model that presents the best performance is the ensemble or combinatorial method of random forest regressor models, with an acceptable prediction range (measured with the root-mean-square-error).

Sectional Leading Edge Vortex Lift and Drag Coefficients of Autorotating Samaras

Autorotating samaras such as Sycamore seeds are capable of descending at exceptionally slow speeds and the secret behind this characteristic is attributed to a flow mechanism known as the leading edge vortex (LEV). A stable LEV is known to increase the maximum lift coefficient attainable at high angles of attack and recent studies of revolving and flapping wings have proposed suitable lift and drag coefficient models to characterise the aerodynamic forces of the LEV. For the samara, however, little has been explored to properly test the suitability of these low-order lift and drag coefficient models in describing the aerodynamic forces produced by the samara. Thus, in this paper, we aim to analyse the use of two proposed aerodynamic models, namely, the normal force and Polhamus models, in describing the sectional aerodynamic lift of a samara that is producing a LEV. Additionally, we aim to quantify the aerodynamic parameters that can describe the lift and drag of the samara for a range of wind speed conditions. To achieve this, the study first examined the samara flight data available in the literature, and from it, the profiles of the lift coefficient curves were investigated. Subsequently, a numerical Blade Element-Momentum model (BEM) of the autorotating samara encompassing different lift profiles was developed and validated against a comprehensive set of samara flight data, which were measured from wind tunnel experiments conducted at the University of Bristol for three different Sycamores. The results indicated that both the normal force and Polhamus lift models combined with the normal force drag can be used to describe the two-dimensional lift characteristics of a samara exhibiting an LEV. However, the normal force model appeared to be more suitable, since the Polhamus relied on many assumptions. The results also revealed that the aerodynamic force parameters can vary with windspeed and with the samara wing characteristics, as well as along the span of the samara wing. Values of the lift curve slope, zero-lift drag coefficient, and maximum lift coefficient are predicted and presented for different samaras. The study also showed that the low-order BEM model was able to generate a good agreement with the experimental measurements in the prediction of both rotational speed and thrust. Such a validated BEM model can be used for the initial design of bio-inspired rotors for micro-air vehicles.

Study of Delay Prediction in the US Airport Network

In modern business, Artificial Intelligence (AI) and Machine Learning (ML) have affected strategy and decision-making positively in the form of predictive modeling. This study aims to use ML and AI to predict arrival flight delays in the United States airport network. Flight delays carry severe social, environmental, and economic impacts. Deploying ML models during the process of operational decision-making can help to reduce the impact of these delays. A literature review and critical appraisal were carried out on previous studies and research relating to flight delay prediction. In the literature review, the datasets used, selected features, selected algorithms, and evaluation tools used in previous studies were analyzed and influenced the decisions made in the methodology for this study. Data for this study comes from two public sets of domestic flight and weather data from 2017. Data are processed and split into training, validation, and testing data. Subsequently, these ML models are evaluated and compared based on performance metrics obtained using the testing data. The predictive model with the best performance (in choosing between logistic regression, random forest, the gradient boosting machine, and feed-forward neural networks) is the gradient boosting machine.

A Self-Supervised Fault Detection for UAV Based on Unbalanced Flight Data Representation Learning and Wavelet Analysis

This paper aims to build a Self-supervised Fault Detection Model for UAVs combined with an Auto-Encoder. With the development of data science, it is imperative to detect UAV faults and improve their safety. Many factors affect the fault of a UAV, such as the voltage of the generator, angle of attack, and position of the rudder surface. A UAV is a typical complex system, and its flight data are typical high-dimensional large sample data sets. In practical applications such as UAV fault detection, the fault data only appear in a small part of the data sets. In this study, representation learning is used to extract the normal features of the flight data and reduce the dimensions of the data. The normal data are used for the training of the Auto-Encoder, and the reconstruction loss is used as the criterion for fault detection. An Improved Auto-Encoder suitable for UAV Flight Data Sets is proposed in this paper. In the Auto-Encoder, we use wavelet analysis to extract the low-frequency signals with different frequencies from the flight data. The Auto-Encoder is used for the feature extraction and reconstruction of the low-frequency signals with different frequencies. To improve the effectiveness of the fault localization at inference, we develop a new fault factor location model, which is based on the reconstruction loss of the Auto-Encoder and edge detection operator. The UAV Flight Data Sets are used for hard-landing detection, and an average accuracy of 91.01% is obtained. Compared with other models, the results suggest that the developed Self-supervised Fault Detection Model for UAVs has better accuracy. Concluding this study, an explanation is provided concerning the proposed model’s good results.

Aerodynamic characterisation of delta wing unmanned aerial vehicle using non-gradient-based estimator

AbstractAerodynamic characterisation from flight testing is an integral subroutine for evaluating a new flight vehicle’s aerodynamic performance, stability and controllability. The estimation of aerodynamic parameters from flight test data has extensively been explored, in the past, using estimation methods such as the equation error method, output error method and filter error method. However, in the current era, non-gradient-based estimation techniques are gaining attention from researchers due to their inherent data-driven optimisation capability to find the global best solution. In this paper, a novel non-gradient-based estimation method is proposed for the aerodynamic characterisation of unmanned aerial vehicles from flight data, which relies on the maximum likelihood method augmented with particle swarm optimisation. Flight data sets of a wing-alone unmanned aerial vehicle are used to demonstrate the capabilities of the proposed method in estimating aerodynamic derivatives. Estimates from the proposed method are corroborated with the wind tunnel test and output error method results. It has been observed that simulated flight vehicle responses using estimated parameters are in good agreement with measured data in most of the manoeuvers considered. Confidence in the estimates of linear and nonlinear aerodynamic parameters is well established with the lower limit of Cramer-Rao bounds, which are minimal. The proposed method also demonstrates good predictability of the quasi-steady stall aerodynamic model by estimating stall characteristic parameters such as aerofoil static stall characteristics parameter, hysteresis time constant and breakpoint. The overall performance of the proposed estimation method is on par with the output error method and is validated with the proof-of-match exercise.

Detailed Analysis of Aircraft Fuel Flow Using Data from Flight Data Recorder

This paper aims to perform accurate prediction of fuel flow (FF) by employing various models: deep learning (DL), random forest (RF), generalized linear model (GLM), and the Eurocontrol Base of Aircraft Data (BADA) model, and to examine the link between FF and different aircraft performance parameters. The flight data set used in this study is obtained from real turbofan engine narrow-body aircraft performing short-distance domestic subsonic flights, containing a total of 2,674 cruise flights between 31 city pairs. Several statistical error analyses are conducted to compare the performance of the models. Root mean square error, mean absolute error, and mean squared error values for DL are calculated to be 0.01, 0.008, and 0.0001, respectively. On the coefficient of determination ( [Formula: see text]) test for validation, the DL model has the highest value of 0.94. Results of the analysis in this paper show that the DL model offers the best ability to predict FF in all statistical tests, which makes it the best-suited model to estimate FF. These findings can provide the means to make more efficient trajectory planning and forecasts in air traffic management, and more accurately predict fuel consumption of aircraft, thereby decreasing emission levels of carbon gases and of various pollutants and airline operating expenses from fuel costs.

Parametric model identification of delta wing UAVs using filter error method augmented with particle swarm optimisation

AbstractFrom arsenal delivery to rescue missions, unmanned aerial vehicles (UAVs) are playing a crucial role in various fields, which brings the need for continuous evolution of system identification techniques to develop sophisticated mathematical models for effective flight control. In this paper, a novel parameter estimation technique based on filter error method (FEM) augmented with particle swarm optimisation (PSO) is developed and implemented to estimate the longitudinal and lateral-directional aerodynamic, stability and control derivatives of fixed-wing UAVs. The FEM used in the estimation technique is based on the steady-state extended Kalman filter, where the maximum likelihood cost function is minimised separately using a randomised solution search algorithm, PSO and the proposed method is termed FEM-PSO. A sufficient number of compatible flight data sets were generated using two cropped delta wing UAVs, namely CDFP and CDRW, which are used to analyse the applicability of the proposed estimation method. A comparison has been made between the parameter estimates obtained using the proposed method and the computationally intensive conventional FEM. It is observed that most of the FEM-PSO estimates are consistent with wind tunnel and conventional FEM estimates. It is also noticed that estimates of crucial aerodynamic derivatives ${C_{{L_\alpha }}},\;{C_{{m_\alpha }}},\;{C_{{Y_\beta }}},\;{C_{{l_\beta }}}$ and ${C_{{n_\beta }}}$ obtained using FEM-PSO are having relative offsets of 2.5%, 1.5%, 6.5%, 3.4% and 7.6% w.r.t. wind tunnel values for CDFP, and 1.4%, 1.9%, 0.1%, 9.6% and 7.5% w.r.t. wind tunnel values for CDRW. Despite having slightly higher Cramer-Rao Lower Bounds of estimated aerodynamic derivatives using the FEM-PSO method, the simulated responses have a relative error of less than 0.10% w.r.t. measured flight data. A proof-of-match exercise is also conducted to ascertain the efficacy of the estimates obtained using the proposed method. The degree of effectiveness of the FEM-PSO method is comparable with conventional FEM.

Meta-Heuristic Solver with Parallel Genetic Algorithm Framework in Airline Crew Scheduling

Airline crew scheduling is a very important part of the operational planning of commercial airlines, but it is a linear integer programming problem with multi-constraints. Traditionally, the airline crew scheduling problem is determined by solving the crew pairing problem (CPP) and the crew rostering problem (CRP), sequentially. In this paper, we propose a new heuristic solver based on the parallel genetic algorithm and an innovative crew scheduling algorithm, which improves traditional crew scheduling by integrating CPP and CRP into a single problem. The innovative scheduling method includes a global heuristic search and an adjustment for flights and crew so as to realize crew scheduling. The parallel genetic algorithm is used to divide the population into multiple threads for parallel calculation and to optimize the randomly generated flight sequence to maximize the number of flights that meet the crew configuration. Compared with the genetic algorithm, CPLEX and Gurobi, it shows high optimization efficiency, with a time reduction of 16.57–85.82%. The experiment shows that our crew utilization ratio is higher than that for traditional solvers, achieving almost 44 flights per month, with good scalability and stability in both 206 and 13,954 flight datasets, and can better manage airline crew scheduling in times of crew scarcity.

On the Performance of Machine Learning Based Flight Delay Prediction – Investigating the Impact of Short-Term Features

People and companies today are connected around the world, which has led to a growing importance of the aviation industry. As flight delays are a big challenge in aviation, machine learning algorithms can be used to forecast those. This paper investigates the prediction of the occurrence of flight arrival delays with three prominent machine learning algorithms for a data set of domestic flights in the USA. The task is regarded as a classification problem. The focus lies on the investigation of the influence of short-term features on the quality of the results. Therefore, three scenarios are created that are characterised by different input feature sets. When forgoing the inclusion of short-term information in order to shift the prediction timing to an early point in time, an accuracy of 69.5% with a recall of 68.2% is achieved. By including information on the delay that the aircraft had on its previous flight, the prediction quality increases slightly. Hence, this is a compromise between the early prediction timing of the first model and the good prediction quality of the third model, where the departure delay of the aircraft is added as an input feature. In this case, an accuracy of 89.9% with a recall of 83.4% is obtained. The desired timing of prediction therefore determines which features to use as inputs since short-term features significantly improve the prediction quality.