Machine learning for helicopter accident analysis using supervised classification: Inference, prediction, and implications

The work is part of a larger effort whose end-objective is to contribute toward a better understanding of helicopter accidents and improving their safety track record. Herein, we extend the domain of application of Machine Learning (ML) to a new topic, namely helicopter accidents. Our objectives are twofold: (1) to benchmark the performance of different classifiers in examining our dataset, and (2) to leverage the best-in-class classifier to identify novel insights for improving helicopter accident analysis and prevention. Following the training and tuning of six different classifiers, we benchmark their predictive performance and identify the Deep Neural Network (DNN) as the best-in-class. We then leverage it to analyze the probability of helicopter accident by controlling for different features in the dataset, namely the number of main rotor blades, number of engines, rotor diameter, and weight. We identify and rank the best- and worst-in-class helicopter configurations in different weight categories, and we discuss the implications of our results. We found, for example, that for light helicopters, larger rotor diameters with increasing number of blades are associated with higher probability of accidents. We also identify regions in the configurations space where twin-engine helicopters are less safe than their single-engine counterparts. Overall, this work demonstrates signifcant opportunities for applying data-driven ML approaches to helicopter accident analysis and how to leverage these tools for extracting value out of datasets for novel insights and safety improvement.