Source rock properties such as kerogen type are critical for reservoir quality evaluation of unconventional oil and gas resources. However, determining the kerogen type is challenging due to the significant formation heterogeneities and poor correlations between geophysical logging data and laboratory geochemical observations. In this paper, a genetic algorithm–driven support vector machine (GA-SVM) method was developed for kerogen type identification based on geological conditions, well logging, and geochemical data. The core idea of the method is, first, to classify the small sample data with the SVM. The logging data (the neutron, sonic, density, gamma ray, resistivity, uranium content, potassium content, and thorium content) and the kerogen type identified by the geochemical crossplot were used as a training data set. Second, the GA was used to optimize the parameters, determine the optimal penalty parameter and kernel function under cross-validation, realize the quantitative characterization of the spatial distribution of kerogen type and interlayer heterogeneity, and evaluate the accuracy of the prediction results. The test results show that the GA-SVM is more accurate and efficient than the grid search algorithm and the traditional method of using crossplot for determining the kerogen type of the lacustrine shale.