Explainable Supervised Learning Models for Aviation Predictions in Australia

Artificial intelligence (AI) has demonstrated success across various industries; however, its adoption in aviation remains limited due to concerns regarding the interpretability of AI models, which often function as black box systems with opaque decision-making processes. Given the safety-critical nature of aviation, the lack of transparency in AI-generated predictions poses significant challenges for industry stakeholders. This study investigates the classification performance of multiple supervised machine learning models and employs SHapley Additive exPlanations (SHAPs) to provide global model explanations, identifying key features that influence decision boundaries. To address the issue of class imbalance in the Australian Transport Safety Bureau (ATSB) dataset, a Variational Autoencoder (VAE) is also employed for data augmentation. A comparative evaluation of four machine learning algorithms is conducted for a three-class classification task:—Support Vector Machine (SVM), Logistic Regression (LR), Random Forest (RF), and a deep neural network (DNN) comprising five hidden layers. The results demonstrate a competitive performance across accuracy, precision, recall, and F1-score metrics, highlighting the effectiveness of explainable AI techniques in enhancing model transparency and fostering trust in AI-driven aviation safety applications.