Catastrophe Duration and Loss Prediction via Natural Language Processing variance

Abstract

Textual information from online news is more timely than insurance claim data during catastrophes, and there is value in using this information to achieve earlier damage estimates. This research used text-based information to predict the duration and severity of catastrophes. We constructed text vectors using Word2Vec and BERT models, then used Random Forest, LightGBM, and XGBoost as learners, all of which showed more satisfactory prediction results. This new approach provides timely warnings of catastrophe severity, which can aid decision making and support appropriate responses.