Machine Learning Methods for Development of a Model of Inter-Relation Between Seismic and Thunderstorm Parameters

In this paper, the hypothesis about existence of a statistical relationship between spatial distribution of lightnings and earthquake magnitudes is verified. The spatial distribution of lightnings is quantified and a model describing inter-relation of lightning and earthquake parameters is constructed with the help of machine learning methods. The model parameters are constructed from lightning location data in the in the region of Kuril-Kamchatka arc. The optimal shape of the feature space is constructed using the principal component method. The model is built using the method of gradient boosting on decision trees. To find an optimal set of hyperparameters, the grid search with cross validation is used. The quality of the model results is evaluated using regression analysis metrics. Model interpretation is constructed using permutation importances. Based on the results of the model evaluation, the model is statistically stable and has a relatively high accuracy, indicatinf existence of an interpretable nonlinear statistical relationship.