An Efficient Causal Structure Learning Algorithm Based on Recursive Simultaneous Equations Model

A new algorithm, the RSEM (Recursive simultaneous equations model) algorithm, is presented for causal structure learning under the LSEM (Linear structural equations model). The algorithm effectively applies recursive simultaneous equations model to causal structure learning. This paper makes two specific contributions. Firstly, under the assumption that knowing the causal order of the variables, we show that recursive simultaneous equations model can be used for causal structure learning under the LSEM regardless of whether the datasets follow multivariate Gaussian distribution. Secondly, the performance of the RSEM algorithm is compared with the state-of-the-art algorithms on 7 networks. Simulation results show that the RSEM algorithm outperforms existing algorithms in terms of time performance, and has a quite high accuracy for thresholds 0.005 and 0.01.