A Recursive Filtering Model For Complete Causal Discovery
Ömer Kırkağaçlıoğlu (Department of Computational Sciences and Engineering, Koc Universtiy, Turkey)
Due to the extra computations needed for causal discovery,several methods that are devised for causal exploration uses feature selection algorithms as filters to identify relevant features. Following this line of work, in this study an algorithm that uses this kind of filtering at each depth of the causal hierarchical tree is presented as a way of eliminating noisy features and updating the training set accordingly to increase the performance and accuracy of the causal exploration process. For the discovery of complete causal structure the proposed method will span the causal graph recursively at depth t eliminating the features that are identified to be in the Markov Blanket from depth t-1. For this study a combination of tools from the Causal Explorer package are utilized both for the feature selection filters and the causal discovery algorithms.