Predicting dangerous seismic activity with Recurrent Neural Networks

In this paper we present a solution to the AAIA'16 Data Mining Challenge. The goal of the challenge was to predict, from multivariate time series data, periods of increased seismic activity which may cause life-threatening accidents in underground coal mines. Our solution is based on Recurrent Neural Network with Long Short-Term Memory cells. It requires almost no feature engineering, which makes it easily applicable to other domains with multivariate time series data. The method achieved the 5th place in the AAIA'16 competition, out of 203 teams.