Abstract: The four-digit EC number contains the enzyme names and the chemical equations on which the enzyme acts. In this study, we created a model that predicts an EC number of optimal enzyme candidates for a chemical reaction. Subsequently, the model was evaluated whether it predicts the correct EC number, using the enzyme reaction data listed in the Kyoto Encyclopedia of Genes and Genomes (KEGG), BRENDA and other literature. We developed a Random Forests (RF) prediction model to predict the subclass and sub-subclass (second and third digits) of the chemical equation belonging to EC 3, which consists of two substrates and two products. First, character data of EC number and chemical equation were obtained from KEGG and converted into values. For quantification, the amounts of changes in 208 descriptors (physical and chemical property values) when substrates change into products were calculated for each chemical equation and 208-dimensional feature vectors of the chemical equations were created. Next, SMOTE was applied to oversample 962 feature vectors to 3100 vectors. Then, descriptor selection was performed for model creation. Forward selection was applied to the RF and 23 descriptors were selected. Parameter tuning resulted in a maximum decision tree depth of 15 and the number of trees of 800. The predictive results of the created model yielded an average F1 score of 0.99 for the test data of KEGG. In addition, the prediction accuracy was currently enough for 12 literature reactions listed in BRENDA.