Machine Learning Models for Storm Damage Prediction in Nova Scotia, Canada

Power loss due to storms is a recurring problem in Nova Scotia (NS) that affects hundreds of thousands of households every year. Accurate prediction of the location and magnitude of power outages caused by storms is crucial for developing effective response plans. This study proposes and tests two machine learning models, a deep neural network and a random forest, for storm damage prediction in NS. Historical weather and damage data are used to train the models, which incorporate multiple input features such as wind speed, gust and direction. Both models are found capable of predicting outages with a mean absolute error of around 0.05 (i.e., an accuracy of 95%), outperforming the currently-used regression based tool. Furthermore, comparison against other machine learning algorithms demonstrates the advantages of the proposed models in terms of accuracy, interpretability, scalability and robustness.