Tsetlin LOB: Realtime Regime Learning and Interpretable Prediction in Financial Limit Orderbooks using Convolutional Tsetlin Machines

We propose a Convolutional Tsetlin Machine approach for the problem of detecting and analysing regime changing in multivariate time series generated by financial limit orderbooks. Such time series comprise of multiple time series exhibiting changing behaviors at different time intervals and contain different nonlinear noise and signal characteristics over time. Learning and predicting these changing regime behaviors has several challenges: (1) failure to take relationships between time series into consideration for discovering regimes in multiple time series; (2) lack of an effective approach that models time-dependent behaviors exhibited by series; (3) difficulties in handling outliers or discontinuities which may be informative. We show that our adapted approach of Convolutional Tsetlin Machines to multivariate time series regime classification enjoys many properties of traditional Tsetlin Machines while providing informative insights to regime detection in limit orderbooks while aiming to overcome these challenges.