Predicting the lithology, lithofacies and reservoir fluid classes of igneous rocks holds significant value in the domains of CO₂ storage and reservoir evaluation. However, no precedent exists for research on the multi-label identification of igneous rocks. This study proposes a multi-label data augmented cascade forest method for the prediction of multilabel lithology, lithofacies and fluid using 9 conventional logging data features of cores collected from the eastern depression of the Liaohe Basin in northeastern China. Data augmentation is performed on an unbalanced multi-label training set using the multi-label synthetic minority over-sampling technique. Sample training is achieved by a multi-label cascade forest consisting of predictive clustering trees. These cascade structures possess adaptive feature selection and layer growth mechanisms. Given the necessity to focus on all possible outcomes and the generalization ability of the method, a simulated well model is built and then compared with 6 typical multi-label learning methods. The outperformance of this method in the evaluation metrics validates its superiority in terms of accuracy and generalization ability. The consistency of the predicted results and geological data of actual wells verifies the reliability of our method. Furthermore, the results show that it can be used as a reliable means of multi-label prediction of igneous lithology, lithofacies and reservoir fluids.