Machine learning models, widely applied in landslide susceptibility assessment due to their powerful feature extraction capabilities, are continuously evolving in their algorithms to address the common issue of low accuracy. The GCNN (group convolutional neural network) model was introduced into landslide susceptibility assessment, and its results were compared with those of various common machine learning models to comprehensively evaluate the adaptability of these models in this field. Taking Hebei Province as the research area, 16 influencing factors were selected from three aspects: triggering factors, pregnant disaster environment, and susceptible bodies. GCNN model and other common machine learning models—CNN (convolutional neural network), Logistic (logistic regression), RF (random forest), and SVM (support vector machine) —were constructed to build corresponding susceptibility assessment models. The research area is divided into four categories of landslide susceptibility zones, and the accuracy of the zoning is comprehensively evaluated. The study indicates that compared with the other four machine learning models, the GCNN model achieves higher scores in various confusion matrix indicators and is more suitable for landslide susceptibility zoning. The resulting zoning of landslide susceptibility is consistent with the actual occurrence of landslide points, indicating a more accurate delineation of landslide-prone areas.