Abstract
A construction method of an abnormality diagnosis model for diagnosing an abnormality of a process, the construction method includes: creating a first regression model that sets a regression coefficient regarding an explanatory variable with a small influence on a response variable to zero by using all operational data in normal times collected in advance; dividing the operational data into a plurality of segments determined in advance and determining an explanatory variable candidate for each of the segments within a range of the explanatory variable used in the first regression model; and creating a second regression model that sets a regression coefficient regarding an explanatory variable candidate with a small influence on a response variable to zero by using the operational data included in the segment.