This letter investigates the utilization of pulse heating and machine learning techniques to overcome the limitations associated with traditional testing methods for metal oxide semiconductor (MOS) gas sensors. These limitations include long-term drift, high power consumption, and challenges in multitasking. Pulsed heating is used to improve long-term stability and significantly reduce power consumption. Three machine learning approaches on top of two models are specially tailored to simultaneously handle gas identification and concentration detection tasks. The experimental results corroborate the robust classification aptitude of all three models and their satisfactory regression accuracy. Moreover, each model offers distinct advantages and can be utilized to meet particular requirements. This letter highlights the potential of pulse heating combined with machine learning to enhance the capabilities of MOS gas sensors.