Gas identification is one of the most important functions of a gas sensor system. To identify gas species from sensing signals without gas flow control such as pumps or mass flow controllers, it is necessary to extract decisive dynamic features from complex sensing signals due to uncontrolled airflow. For that purpose, various analysis methods using system identification techniques have been proposed, whereas a method that is not affected by a gas input pattern has been demanded to enhance the robustness of gas identification. Here we develop a novel gas identification protocol based on a transfer function ratio (TFR) that is intrinsically independent of a gas input pattern. By combining the protocol with MEMS-based sensors-Membrane-type Surface stress Sensors (MSS), we have realized gas identification with a free-hand measurement, in which one can simply hold a small sensor chip near samples. From sensing signals obtained through the free-hand measurement, we have developed highly accurate machine learning models that can identify odors of spices and herbs as well as solvent vapors. Since no bulky gas flow control units are required, this protocol will expand the applicability of gas sensors to portable electronics, leading to practical artificial olfaction.