Method and device for estimating value to be estimated associated with specimen

A method of estimating a specific estimation target value associated with a specific chemical component of an unknown specimen based on a result of machine learning learned from an output of an array of a plurality of chemical sensors, wherein chemical sensors in the array of the plurality of chemical sensors responds to at least one of the plurality of chemical components comprising the specific chemical component of the unknown specimen differently from each other, the method comprising:

• obtaining signal values at a plurality of selected time points in at least one cycle of each periodic variation of the output when an amount of a single specimen given to the plurality of chemical sensors is periodically varied, the single specimen being one of a plurality of known specimens having known values of the specific estimation target;
• extracting a plurality of parameters which represent a characteristic of a response of the plurality of chemical sensors from the obtained signal values at the plurality of selected time points for each of the plurality of known specimens;
• performing machine learning to obtain a relationship between the values of the specific estimation target and the characteristic of the response of the plurality of chemical sensors, based on differences in the plurality of parameters, without using a saturated value of the output;
• obtaining an output from the array of the plurality of chemical sensors for the unknown specimen;
• and
• estimating the specific estimation target value of the unknown specimen using the obtained relationship.