A natural gas (NG) odorization system requires a continuous monitoring system in addition to the proper injection setup. As other sulfur based compounds such as hydrogen sulfide can also coexist in the environment, one of the challenges associated with monitoring the odorants is the selectivity. In this thesis, first, microfluidic gas sensors are modified to enhance their selectivity. In our first study, the effect of microchannel geometry modification (reduction of width) is studied through a computational parametric approach. The results show an average improvement of 93.44% and 60.1% in selectivity toward polar and nonpolar VOCs, respectively. In the next step, the surface of the microchannel is modified with graphene quantum dots, which has a two-fold effect:​ (i) increasing the surface area, and (ii) adding functional groups. The experimental results of this step show an average improvement of 98.72% and 86.61% in the sensor’s selectivity toward polar and nonpolar VOCs, respectively. In our second study, optimized cylindrical microfeatures are embedded into the microchannel. The microchannel is coated with graphene oxide, to increase the surface to volume ratio by introducing nanofeatures to the surfaces. The changes in the sensor response are compared to a plain microfluidic gas sensor, the results show an average of 64.42% and 120.91% improvement in the selectivity of the sensor with microfeatures and both nano- and microfeatures, respectively (toward VOCs). In the last part of this project, a portable device integrated with an array of five different sensors and a machine learning model are developed for the selective detection and prediction of NG odorant:​ a mixture of tert-butyl mercaptan and methyl ethyl sulfide for a concentration range of 1 ppm to 10 ppm. The device shows the sensitivity and selectivity indicator of 0.36 (1 / ppm), and 28.59 toward odorant respectively. In the next step, we achieved a 98.75% classification accuracy between NG odorant and hydrogen sulfide. The overall Mean Square Error (MSE) and R² regression model of 0.50 and 95.16% have also been achieved. These results indicate that the developed device and the prediction model have promising applications for selective monitoring of the NG odorant.