This paper presents the prototype of a wireless gas sensor system with real-time monitoring interface. A learning-based performance regulation scheme is proposed to improve the system efficiency. The scheme can adjust the pattern recognition, mode control and prediction of successive approximation register analog-to-digital converter (SAR ADC) results with neural network algorithm. The ROIC is supported 8-channel gas sensors and multi-mode as monitoring mode and precision mode. In order to optimize the power consumption of precision mode further, auto controlled correlated double sampling (CDS) zooming is proposed in the ROIC. Thus, the system can optimize the required resolution and power consumption depending on the critical level of each gas type and concentration. The prototype ROIC was fabricated with CMOS technology and 15.8x efficiency improvement of the system were verified experimentally through real gas sensing.