The identification of fluid types has always been the focus of oil and gas field exploration and development research. At this stage, a large amount of CO 2 has been found in many basins during exploration and development, which greatly affects the accuracy of reservoir understanding and evaluation, so it is very important to accurately identify the fluid type of the CO 2 -bearing reservoirs. However, methods utilizing multiple logging data are greatly affected by physical changes in the formation, resulting in methods that are only applicable to the area and layer under study, are poorly generalized, and require multiple instruments and experimental support. Existing nuclear logging methods that primarily utilize logging curve stacking and intersection map methods fail to take full advantage of logging. In this study, taking advantage of the fact that the neutron gamma technique of nuclear logging measurement can provide multiparameter information with the characteristics of machine learning to deal with multidimensional data, comparing the classification results of different ware learning methods under different classification strategies and selecting a method of identifying fluids in logging while drilling was based on the idea of dichotomy and the use of the support vector machine as a meta-model. This solved the problem of identifying fluids in the CO 2 -bearing reservoirs, providing new ideas for the design and fabrication of logging instruments. These instruments can assist with the exploration and development of oil and gas fields and has a broad application prospect in CO 2 -EOR and CCUS.