3.1.1 AI-driven emission reduction and fuel efficiency

This theme covers studies where AI and ML methods are used to reduce carbon emissions, optimise fuel consumption and improve environmental performance in aviation. A total of 18 studies were included in the review under this theme. These studies cover various applications such as the optimisation of SAF, hybrid engine modeling, fuel load planning, engine emission prediction and preventing unnecessary fuel consumption caused by air traffic operations. AI applications have demonstrated a significant effect on enhancing sustainability by improving operational efficiency and supporting the ICAO’s long-term CO₂ reduction goals. ML-supported predictive analytics play a critical role in reducing carbon emissions in production processes and offer decision-support systems that enhance energy efficiency through real-time monitoring and anomaly detection [Reference Ojadi, Onukwulu, Odionu and Owulade19]. This approach sheds light on similar processes used in aviation and highlights the general potential of AI-based sustainability strategies. The artificial neural network-based model developed by Kosir et al. [Reference Kosir, Heyne and Graham20] predicts the material compatibility of SAF blends within aircraft fuel systems and enables the development of fuel types optimised for both safety and environmental performance. Similarly, Oh et al. [Reference Oh, Oldani, Solecki and Lee21] developed deep learning models that included uncertainty calculations to predict flash points of sustainable fuels. These studies suggest that AI can be used as a reliable tool, especially in SAF certification processes. Additionally, Wang et al. [Reference Wang, Xue, Wu and Yan22] demonstrated that fuel consumption predictions and optimisation based on the QARQAR data provided an average fuel saving of 3.67%. These models consider ICAO-recommended reserve fuel policies and contribute to strategies that reduce the environmental impact while maintaining operational safety. The power prediction-based model predictive control (P2MPC) system developed by Wei et al. [Reference Wei, Ma, Xiang and Liu23] provided fuel efficiency and emission reduction by controlling the battery status and EGT in hybrid land-air vehicles. This represents a critical AI application in the context of the anticipated proliferation of electric and hybrid aircraft. Sun et al. [Reference Sun, Zhang and Su24] analysed the final demand-based emission intensity of the transportation sector on a more macroscale and showed that air transportation increases the indirect climate burden in developed countries. AI-supported data analysis techniques provide significant advantages in such analyses. Similarly, Miller et al. [Reference Miller, Durlik, Kostecka, Łobodzińska and Matuszak25] emphasised the transformative potential of AI technologies in increasing energy efficiency and reducing greenhouse gas emissions in the transportation sector. The study states that AI-supported solutions in flight route optimisation, management of alternative fuel systems and fleet planning contribute to reducing the carbon footprint in aviation. AI facilitates production, distribution and performance optimisation in hydrogen and biofuel-based systems by enabling process modeling, yield prediction and energy-flow optimisation functions that accelerate SAF integration into existing aviation fuel infrastructures [Reference Miller, Durlik, Kostecka, Łobodzińska and Matuszak25]. In the literature, it is emphasised in many sources that such AI applications play a critical role in achieving sustainability goals. Likewise, the ICAO’s Environmental Report [2] emphasised AI as a technology that should be integrated into carbon reduction strategies. In this context, it is seen that AI technologies in the aviation sector offer multidimensional contributions to achieving sustainability goals not only at the operational level but also at the strategic, environmental and infrastructural levels. The studies examined reveal the potential of AI to increase environmental performance, especially fuel efficiency and emission reduction, with concrete applications.

3.1.2 Aircraft maintenance and reliability with AI support

In the aviation industry, maintenance operations play a critical role in ensuring flight safety and operational continuity. Particularly for airline operators functioning under high safety requirements and cost pressures, optimising aircraft maintenance processes is important [Reference Ahmadi, Söderholm and Kumar26]. While traditional maintenance approaches are mostly based on time-based or reactive strategies, in recent years, AI-based solutions have transformed the sector by making these processes data-driven, predictive and predictive [Reference Pınar27]. AI technologies such as ML, robotics, Internet of Things (IoT), computer vision and natural language processing have been integrated into maintenance processes, enhancing both operational efficiency and technical reliability [Reference Amit28]. For instance, Rolls Royce’s Intelligent Borescope system reduces engine inspection time by 75%, while Lufthansa Technik’s Condition Analytics solution analyses sensor data to predict maintenance needs in advance [Reference Lu29]. These systems not only detect current faults but also contribute to the optimisation of maintenance planning by making future-oriented predictions based on historical usage data and failure trends. These thematic findings are consistent with the studies analysed in this review. For instance, Kabashkin et al. [Reference Kabashkin, Perekrestov, Tyncherov, Shoshin and Susanin30] demonstrated that AI-supported maintenance optimisation in unmanned aerial vehicles (UAVs) led to a 20% cost reduction. In addition, other studies on this theme emphasise various technical dimensions regarding the applications of AI in maintenance processes. For example, Apostolidis et al. [Reference Apostolidis, Bouriquet and Stamoulis31] developed a solution for the damage detection of engine components using neural network-assisted image processing algorithms. Plastropoulos et al. [Reference Plastropoulos, Bardis, Yazigi, Avdelidis and Droznika32] emphasised the role of AI-driven data collection and predictive analytics within the context of ‘smart hangars’, highlighting how technologies such as digital twins, robotics and IoT can optimise aircraft maintenance processes, enhance safety and contribute to sustainability in line with Industry 5.0 principles. In studies conducted within the Turkish context, Kilic et al. [Reference Kilic, Villareal-Valderrama, Ayar, Ekici, Amezquita-Brooks and Karakoc33] demonstrated the applicability of deep learning-based models, particularly LSTM networks, for forecasting micro turbojet engine performance, highlighting their potential to improve maintenance efficiency and support environmental sustainability in local aviation practices. Furthermore, Mohammadi et al. [Reference Mohammadi, Rahmanian, Sattarpanah Karganroudi and Adda34] highlighted the effectiveness of transformer-based deep learning models, particularly DeiT, for early aero-engine defect detection under limited data conditions, demonstrating their potential to enhance predictive maintenance strategies and support sustainability in aircraft operations. By reducing unscheduled maintenance events, extending component lifespans and minimising resource waste, AI-driven predictive maintenance contributes directly to the environmental and economic dimensions of sustainability in aviation.

3.1.3 Airport and infrastructure sustainability

Consistent with the UN’s 2030 Sustainable Development Goals, AI-supported airport operations have increasingly focused on optimising energy consumption, waste management and traffic flow efficiency to reduce environmental impact [Reference Schulte-Sasse35]. Integrating AI into airport operations enables the optimisation of numerous processes, ranging from apron management and passenger flow to baggage logistics and terminal energy management. In the literature, the gains obtained through the integration of AI into airport infrastructure are gathered under the following headings: operational efficiency, environmental sustainability, energy savings, improvement of passenger experience and strengthening of cybersecurity [36, 37]. For example, passenger flows within the terminal are analysed with camera and sensor data, reducing congestion at security, passport control and boarding gates, thus increasing passenger satisfaction. AI-based biometric systems and facial recognition technologies accelerate identity verification processes and increase the accuracy of security protocols. Moreover, AI-supported systems in terminal energy management optimise lighting, HVAC (heating, ventilation, and air conditioning) and overall energy consumption; AI algorithms predict peak usage hours and dynamically adjust operations accordingly [37]. These applications contribute to environmentally friendly air transportation by reducing airports’ carbon emissions. The studies analysed within this theme also confirm the developments observed in the literature. For instance, Ramakrishnan et al. [Reference Ramakrishnan, Seshadri, Liu, Zhang, Yu and Gou38] evaluated the green performance of airports worldwide using explainable semi-supervised learning models. Their model offers sustainability-enhancing strategies to airport operators based on environmental parameters such as S1/S3-type greenhouse gas emissions and energy usage. Similarly, a systematic review by Selvam and Al-Humairi [Reference Selvam and Al-Humairi39] emphasised the contribution of IoT- and AI-based weather monitoring systems to sustainable airport operations. The use of these systems as decision-support tools in air traffic management helps reduce weather-related operational risks and enhances energy efficiency. The integration of AI into airport and infrastructure sustainability is driving a multi-dimensional transformation that not only increases operational efficiency but also increases environmental awareness. In light of the increasing air traffic, energy costs and environmental regulations, the comprehensive adoption of these technologies has become a critical strategic priority to both meet climate goals and improve passenger satisfaction.

3.1.4 AI applications in aviation education, decision support and policy making

As part of the digital transformation process, the aviation sector has begun to effectively use AI technologies not only in operational processes but also in education, decision support and policy making. In the context of education, large language models (LLM) and AI-supported assistants facilitate students’ access to information, improve programming skills and personalise learning processes [Reference Wandelt, Sun and Zhang40]. The study by Wandelt and colleagues [Reference Wandelt, Sun and Zhang40] empirically demonstrated the potential of ChatGPT to enhance student performance in aviation management education, while also highlighting the need for critical evaluation in terms of accuracy, reliability and contextual alignment. Within the scope of decision support systems, Klophaus [Reference Klophaus41] conducted a study utilising AI to perform SWOT analyses on five emerging aviation technologies (Urban Air Mobility, UAVs, electric aircraft, supersonic flights and SAFs). The study emphasises that AI’s capacity to produce analytical outputs can provide meaningful contributions to decision-making processes when complemented by expert opinions. This indicates that AI systems can play a supporting role in aviation strategies. AI-supported education and decision-making tools also foster sustainability by promoting safety culture, energy awareness and data-driven policymaking aligned with sustainable operational standards. When evaluated in this context, it is seen that AI-supported systems have significant potential in aviation education, decision support mechanisms and policy production. Nevertheless, the reliability, ethical implementation and human-centred design of such technologies must be carefully addressed. The principles of human oversight, transparency, and accountability should be upheld [Reference Schulte-Sasse35]. Thus, AI will continue to contribute to the digitalisation and sustainability goals of the aviation sector not only at the operational level but also in the areas of governance and education.

Collectively, these four themes demonstrate that AI technologies are being increasingly integrated into aviation operations to enhance sustainability, safety, and efficiency. While the corpus of 27 studies provides a structured overview of emerging trends, its limited size constrains the generalisation of sector-wide conclusions and should be interpreted as exploratory rather than exhaustive.

3.3.1 AI-based safety prediction: random forest (RF)

AI offers significant contributions to enhancing operational safety in aviation. In particular, RF, a supervised learning algorithm, is widely used to predict safety-critical events by analysing flight data. This method enables the identification of patterns that are difficult to detect using traditional inspection techniques, thus allowing risky situations to be anticipated in advance [Reference Breiman42, Reference Li, Wen and Su43]. In this context, AI-powered predictive maintenance (PdM) applications are rapidly gaining importance in the aviation industry. With systems based on historical maintenance records, IoT-based sensor data and visual inspections, it is possible to predict mechanical failures with high accuracy before they occur. This not only reduces maintenance costs but also increases system reliability [Reference Rubinstein44]. An example of this innovative approach is the camera-based PdM system developed by the Israel-based company OdysightAI. This system was developed following the 2017 crash of an Apache helicopter caused by a failure in a flight control rod that was not monitored by existing maintenance systems. Using micro-cameras and specialised AI models, the system continuously monitors critical components and provides early warnings. This development adds a visual layer to traditional prognostics and health monitoring (PHM) systems for components that are otherwise not covered [Reference Rubinstein44]. The system is currently deployed in IDF Apache helicopters and is being tested for broader integration in other rotorcraft platforms and under extreme environmental conditions in collaboration with NASA. Analyses show that such systems provide a 40% cost saving compared to reactive maintenance methods and 8%–12% compared to preventive maintenance systems. AI’s contribution to aviation safety is not limited to individual aircraft systems. Ricci [Reference Ricci45] states that commercial aviation is still the safest form of transportation and that according to International Air Transport Association (IATA) data, only 0.16 accidents occur per million flight segments. However, increasing air traffic density and inadequacies in critical workforce areas such as pilots, air traffic controllers and maintenance technicians constitute the main safety challenges facing the aviation sector. At this point, ML techniques such as RF play an important role in operational decision support systems by increasing situational awareness and reducing the cognitive load of maintenance or flight crews. Moreover, in systems developed by Airbus, such as Wayfinder and Optimate, RF-based models are actively employed in advanced functions, including autonomous flight control and 4D trajectory management. These systems aim to conduct operations both safely and efficiently by analysing multivariate inputs such as weather, traffic density, fuel consumption and even pilot attention distributions in real-time [Reference Ricci45]. In this context, it can be said that AI-based modeling and prediction tools offer a safety architecture that covers not only maintenance processes but also holistic flight systems.

3.3.2 Aircraft routing and performance forecasting: XGBoost

Flight routing and performance prediction hold strategic importance for airline operations in terms of both operational efficiency and energy management. Traditional route planning approaches typically rely on fixed rules and historical data, failing to adequately account for dynamic variables such as weather conditions, traffic density and real-time system behaviours. In this regard, powerful ML algorithms like extreme gradient boosting (XGBoost) offer significant advantages in predicting flight performance and routing decisions with high accuracy [Reference Chen and Guestrin46, Reference Rajendran and Sundararajan47]. XGBoost is particularly prominent in predicting microturbine engine performance and modeling parameters such as fuel consumption and aerodynamic drag during flight [Reference Paraschos, Trimble, Bhargava, Klingler and Nicolai48]. This algorithm predicts the behaviour of aircraft systems under certain operational conditions by learning from historical flight data. It enables the development of more efficient, environmentally sustainable and time-optimised flight plans in line with these predictions [Reference Li, Wen and Su43]. Moreover, XGBoost has demonstrated strong predictive capabilities in airport ground operations. In a study by Wang [Reference Wang, Xue, Wu and Yan22], the taxi-in time of arriving aircraft was predicted at an airport located in central-southern China. In the study, 15 affecting factors were determined, the importance of these factors was ranked by the RF method, and then the XGBoost-based prediction model was applied. The model, which was evaluated with criteria such as mean absolute error (MAE), mean square error (MSE) and root mean square error (RMSE), showed 92.78% accurate prediction success within ±5 minutes when compared to linear regression, support vector machines, and artificial neural networks. Paraschos [Reference Paraschos, Trimble, Bhargava, Klingler and Nicolai48] supported the success of XGBoost in flight time estimation with visual analysis. In the study, the relationship between the actual flight times and the times estimated by the XGBoost model was examined with a scatter plot, and the model’s predictive ability was revealed with high accuracy. Each data point corresponds to a specific flight, with the actual values on the horizontal axis and the predicted values on the vertical axis. This graphical approach provides important clues in terms of the model’s stability and predictive success. In addition, another study by Shengyuan et al. [Reference Shengyuan, Xinghua, Guangdong, Wu and Zhao49] showed that XGBoost is also used innovatively in the context of air-rail integrated transportation networks. In this study, a heuristic path planning algorithm (XGB-HPPA) based on XGBoost is proposed for the prediction of transfer stations in time-expanded graphs (TEG) with high data density. The model first reduces the complexity of the network by eliminating inefficient transfer nodes and greatly speeds up the processing time of path-planning algorithms. Comparative analyses show that XGB-HPPA both increase the number of valid solutions and obtains results that converge to the optimal solution. Such applications reveal that XGBoost can be effectively used not only in in-flight system predictions but also in macro-level multimodal transportation planning.

3.3.3 Emission reduction: generative adversarial networks (GAN)

Given that aviation accounts for approximately 2.5% of global greenhouse gas emissions [50], accurate emission modeling has become a priority area for sustainability-focused innovation. In this context, AI algorithms have become a critical tool, especially in scenario simulations aimed at modeling, estimating and reducing carbon emissions [Reference Adewale, Ene, Ogunbayo and Aigbavboa51]. In particular, GAN algorithms provide significant benefits in synthesising low-quality or incomplete environmental data, determining emission sources and testing operational scenarios. Zhang et al. [Reference Zhang, Rong, Liu and Yang52] stated that these algorithms can be used to estimate carbon emissions, especially at seaports and airports; GAN-based data enhancement methods provide more effective results than traditional sensor data in accurately modeling port-based emissions. The same study emphasises that software development processes should be systematically structured from the conceptual design to the final prototype to increase the accuracy of emission analyses. Similarly, a study developed by Zhou [Reference Zhou53] demonstrated significant improvements in accuracy, F1-score and strategy adaptation capabilities in carbon emission prediction through optimisations made on the convolutional neural networks (CNN) architecture. This model was equipped with multi-scale feature extraction, attention mechanisms and advanced optimisation algorithms and was tested on datasets related to fuel use and the carbon footprint of airline companies. The findings showed that the proposed model has much higher accuracy and response time than other traditional AI architectures, and its performance is further improved, especially in environments with large data volumes [Reference Zhou53]. On the other hand, a study conducted by Kosir et al. [Reference Kosir, Heyne and Graham20] revealed that GAN algorithms successfully predicted the volumetric swelling effects on aircraft fuel systems in contact with SAF. In this way, the environmental performance was increased by preserving material compatibility and optimum fuel mixtures could be created in terms of energy efficiency. The ability of GANs to model such physical interactions enables a much more comprehensive and dynamic structure compared to traditional statistical modeling. All these examples demonstrate the multifaceted contributions of AI in emission reduction. The use of models such as GAN, CNN, temporal convolutional network (TCN) and graph convolutional network (GCN) in both direct carbon estimations and operational improvement processes enables the design of environmental sustainability strategies based on scientific foundations. Thus, airlines are empowered to make more proactive and data-driven decisions, not only to comply with regulations but also to reduce operational costs and fulfill their environmental responsibilities.

3.3.4 Training and policies: transfer learning

The integration of AI technologies into aviation training processes and policy development increases the quality of learning in the sector and strengthens strategic decision-making processes. In particular, transfer learning allows models developed for use in different training scenarios to be adapted to other training environments, shortening learning times and increasing performance. For instance, transfer learning models developed for simulator environments in aviation training accelerate pilots’ adaptation to emergency scenarios and significantly improve the effectiveness of such training processes [Reference Savaş, Özdemir and Esen54, Reference Cao, Zhang and Feng55]. Transfer learning enables the adaptation of pre-trained models to specific aviation datasets, reducing the computational demand and training costs, which enhances energy efficiency and resource sustainability in AI model development. In addition, the integration of large language models (LLMs) such as ChatGPT into educational processes supports the faster acquisition of theoretical knowledge and facilitates the comprehension of complex aviation terminology for students [Reference Liu56]. These systems promote interactive and dynamic learning in training environments, thereby enhancing the retention and effectiveness of conceptual understanding. In terms of policy development processes, AI-supported decision-making mechanisms are of critical importance. In particular, AI-based data analytics methods help to better plan and manage operational and strategic decisions in aviation management. The detailed data analysis and predictive capabilities provided by the AI models make it possible to comprehensively evaluate the basic dimensions of air transportation policies, such as sustainability, economic efficiency and safety [Reference Kabashkin, Perekrestov, Tyncherov, Shoshin and Susanin30, Reference Yıldız and Çulha57]. For example, ML algorithms used in decision support systems play an active role in the planning of airspace management and air traffic policies, reducing the cognitive load of air traffic controllers and increasing the level of safety [Reference Milićević58]. As a result, the integration of AI methods into education and policy development processes accelerates the future transformation of the aviation sector and contributes to the creation of a more efficient, safe and sustainable aviation ecosystem.

3.3.5 Aircraft maintenance: hybrid ML

Aircraft maintenance processes are one of the key components of ensuring safety and continuity in aviation operations. Hybrid ML approaches have recently come to the forefront to increase the effectiveness of maintenance activities and prevent operational disruptions. Hybrid models combine different types of algorithms to increase accuracy, early fault detection and predictability compared to the models used alone. Kaynak and Ervural [Reference Kaynak and Ervural59] emphasised that the hybrid ARIMA-ANN model outperformed the classical linear and nonlinear methods in predicting machinery failures. These hybrid models can effectively identify both linear and nonlinear patterns in real-world data, significantly reducing unplanned downtimes. This approach is noted to improve operational efficiency and lower maintenance costs, thereby supporting the competitive advantage of airlines. Jasra et al. [Reference Jasra, Valentino, Muscat and Camilleri60] recommended the use of the local outlier factor (LOF) algorithm together with hybrid statistical methods for anomaly detection in-flight data. This method not only detects anomalies occurring during the flight process but also increases flight safety by quantifying the degree of anomalies. Thus, detailed analysis is provided even during short periods when the flight exhibits abnormal behaviour. This method significantly supports maintenance and safety management processes by reducing human intervention in flight data analysis. Paredes et al. [Reference Paredes, Chávez, Isa-Jara and Vargas61] developed a hybrid approach combining supervised learning (MLP) and reinforcement learning (Q-learning) algorithms to estimate the remaining useful life (RUL) of aircraft engines. The method was found to offer 15% higher accuracy compared to approaches using only supervised learning algorithms and 4% more accuracy than other hybrid methods. It provided an early warning approximately 17 cycles before failure, effectively preventing unexpected machinery downtimes and reducing maintenance costs. In addition to these studies, hybrid models effectively integrate data from different sources such as sensor data, historical maintenance records, and real-time aircraft component status data to predict potential failures more accurately [Reference Kabashkin, Perekrestov, Tyncherov, Shoshin and Susanin30–Reference Plastropoulos, Bardis, Yazigi, Avdelidis and Droznika32, Reference Mohammadi, Rahmanian, Sattarpanah Karganroudi and Adda34]. Furthermore, the integration of computer vision, transfer learning and digital twin technologies into hybrid methods significantly enhances both operational and financial outcomes in maintenance processes [Reference Kabashkin, Perekrestov, Tyncherov, Shoshin and Susanin30, Reference Rubinstein44]. As a result, hybrid ML approaches increase the accuracy of failure predictions in aircraft maintenance processes and provide significant advantages in terms of operational continuity and safety by reducing maintenance costs.

3.3.6 Flight optimisation: long short-term memory (LTSM)

Optimisation of flight operations is a critical goal in the aviation industry in terms of both reducing operational costs and increasing sustainability. In this context, long short-term memory (LSTM) models, which can capture long-term dependencies in time series data, are frequently preferred in-flight optimisation processes. The effectiveness of LSTM models has been confirmed by academic studies conducted at various stages of flight operations. For instance, Qu et al. [Reference Qu, Xiao, Yang and Xie62] developed an Att-Conv-LSTM model that integrates temporal and spatial data to predict flight delays and found that it reduced prediction errors by 11.41% compared to conventional LSTM models. Similarly, Yousefzadeh Aghdam et al. [Reference Yousefzadeh Aghdam, Kamel Tabbakh, Mahdavi Chabok and Kheyrabadi63] stated that bidirectional LSTM models (Bi-LSTM) in air traffic management optimisation, when used together with extreme learning machines (ELM), allow the development of more accurate and faster decision support systems. Xiong et al. [Reference Xiong, Zou, Wan, Sun and Yu64] emphasised the importance of the DMPSO-LSTM model, which they developed based on QAR data, for the accurate prediction of fuel consumption at different stages of flight (climb, cruise and descent). This study increased the quality of data preprocessing by using the adaptive noise ensemble empirical mode decomposition (CEEMDAN) method, and then significantly improved the prediction performance of the LSTM model by using dynamic multi-dimensional particle swarm optimisation (DMPSO). The model results showed a decrease of more than 40% in the MAE value, more than 38% in the RMSE value, and more than 6% in R² value in the climb segment. Similar performance increases were obtained in the cruise and descent segments. The integration of such advanced models into flight operations offers three main advantages to airlines: (1) it provides the most appropriate balance between payload and fuel costs by quantitatively determining the load-fuel relationship; (2) it helps determine the most suitable flight profiles with more accurate predictions at different phases of flight (climb, cruise and descent); and (3) high-accuracy fuel consumption estimates support optimal fuel loading decisions by minimising safety margins [Reference Xiong, Zou, Wan, Sun and Yu64].