Gas identification plays a critical role in characterizing our (chemical) environment. It allows to warn of hazardous gases and may help to diagnose medical conditions. Miniaturized gas sensors, and especially those based on chemiresistive detection mechanisms, have gained rapid development and commercialization in the past decades due to their numerous advantageous characteristics, such as simple fabrication, easy operation, high sensitivity, ability to detect a wide range of gases, and compatibility with miniaturization as well as integration for portable applica-tions. However, they suffer from a remarkable limitation, namely their low selec-tivity. Recently, machine learning-driven approaches to enhance the selectivity of gas sensors have attracted considerable interest in the community of gas sensors, increasing the analyte gas identification ability. In this chapter, firstly, we introduce the general approaches to enhance the selectivity of gas sensors implemented by machine learning techniques, which consists of the architecture scheme design of gas sensors (sensor array and single sensor architecture), the selection of gas sensing response features (steady-state feature and transient-state feature), and the utilization of gas sensing signal modulation techniques (sensing materials modulation, concen-tration modulation, and temperature modulation). Afterward, a specific application case using a machine learning-enabled smart gas sensor for the identification of industrial gases (PH₃ and NH₃) is presented, which is based on a single-channel device and utilizes multiple transient features of the response. We believe machine learning in combination with efficient sensing signal modulation techniques could be a feasible way to gain the gas identification capability of gas sensors.