With the increasing importance of conservation and identification of antiquities, there are numerous methods for the identification of antiquities. In order to realise the analysis and identification of the components of ancient glass objects, this paper establishes a random forest identification model based on the original data, random forest feature extraction and the maximum correlation minimum redundancy algorithm (mRMR), and compares and analyses its effectiveness on various datasets. The mrmr algorithm-based random forest discrimination model achieved a combined accuracy and ROC curve coverage area (AUC value) of 1 on the unweathered dataset, which was close to a perfect model with high confidence in predicting the discrimination results; while the AUC value of the repeatedly trained model was not stable enough on the weathered dataset. Thus, this paper proposes a mrmr-based random forest-based discrimination model, which achieves an optimal AUC value of 0.9393 when the number of decision trees is 30 and the maximum number of splits is 10.