METHOD, DEVICE AND SYSTEM FOR ESTIMATING CAUSAL RELATION BETWEEN OBSERVATION VARIABLES

摘要

To provide a method, device and program for estimating a causal relation between observation variables.SOLUTION: The method involves determining a mixed causal relation objective function suitable for continuous and discrete observation variables in response to the event that the observation data of mixed observation variables is received. The mixed causal relation objective function includes a causal relation objective function of continuous observation variables and a causal relation objective function of discrete observation variables, with a fitting conflict adjusted on the basis of the weight coefficient of observation variables. Then, an optimum solution to the mixed causal relation objective function is obtained by a mixed sparse causal inference which is suitable for both the continuous and the discrete observation variables using mixed observation data under constraints of a directed acrylic graph so as to estimate a causal relation between observation variables.EFFECT: This method is suitable for the causal relation estimation of mixed observation data, and as a causal network structure is not so much sensitive to an estimation error of observation variables, it is possible to obtain an accurate causal relation.SELECTED DRAWING: Figure 2