Machine Learning-Based Turbulence-Risk Prediction Method for the Safe Operation of Aircrafts

Abstract Customer comfort is an important requirement for airlines, and avoiding and mitigating aircraft shaking have always been crucial in this regard. In particular, managing aircraft operations during turbulence is a major issue for airlines. We propose a method for predicting the occurrence of turbulence to support the safe and comfortable operation of aircrafts. Our method integrates meteorological data from Japan and turbulence information provided by Fuji Dream Airlines. Because turbulence occurs rarely, we define a risk cluster that includes turbulence observation data and use it as turbulence training data. Hence, we first estimated the risk cluster, then performed a principal component analysis (PCA) on meteorological data to obtain a projection matrix \(W\) for reducing data dimensions. Using the turbulence-occurrence indicator and the meteorological data coordinates linear transformed by \(W\), we calculated the risk cluster using the k-means method which, in turn, was used in conjunction with support vector classification (SVC) to predict the turbulence-risk dates based on meteorological data from 2019. The results revealed that the days with turbulence risks were accurately identified from the meteorological data; thus, we believe that this method can help support the safe operation of aircrafts. Furthermore, we believe this study will lead to the development of human resources by providing a guide for making safety decisions through the effective use of aviation data.