Application of Data-Based Artificial Intelligence in the Aviation Industry: A Conceptual-Analytic Review of Machine Learning and Deep Learning Methods

Abstract

Abstract: The aviation industry, as a complex and safety-sensitive technical-operational system, faces a huge volume of heterogeneous data, including flight time series, aircraft and engine health data, spatial and temporal air traffic data, meteorological data, textual safety and repair reports, and visual inspection data. This article provides a conceptual-analytical review that aims to explain the logic of "data-driven artificial intelligence" in aviation and describe how machine learning and deep learning methods can be purposefully utilized to produce "operational knowledge" and "actionable decisions." The present review approach, rather than simply comparing algorithms, focuses on the “problem-data-model-output” mapping framework and suggests that the choice of analytical method should be a function of the type of operational problem (prediction, anomaly detection, classification, image analysis, and sequential decision making), the nature of the available data, and the requirements of industrial deployment. The results of the review indicate that classical machine learning methods have greater advantages in more structured problems requiring interpretability, and deep learning models have greater advantages in large/complex or unstructured data (images, text, long time series). However, successful transition from a research environment to an operational environment faces challenges such as lack of labeled data, class imbalance and rare events, changing data scope, need for interpretability, and regulatory constraints. Finally, the paper suggests future directions in the form of multi-source learning, robust and adaptive learning, secure reinforcement learning, human-in-the-loop, and certification frameworks for sustainable deployment of AI in aviation.