Classification-based causality detection in time series

Abstract

Brain effective connectivity aims to detect causal interactions between distinct brain units and it can be studied through the analysis of magneto/electroencephalography (M/EEG) signals. Methods to evaluate effective connectivity belong to the large body of literature related to detecting causal interactions between multivariate autoregressive (MAR) data, a field of signal processing. Here, we reformulate the problem of causality detection as a supervised learning task and we propose a classification-based approach for it. Our solution takes advantage of the MAR model by generating a labeled data set that contains trials of multivariate signals for each possible configuration of causal interactions. Through the definition of a proper feature space, a classifier is trained to identify the causality structure within each trial. As evidence of the efficacy of the proposed method, we report both the cross-validated results and the details of our submission to the causality detection competition of Biomag2014, where the method reached the 2nd place.