Diabetes mellitus is a chronic disease characterized by elevated blood glucose levels. Individuals with diabetes are at increased risk of developing renal, neurological, cardiovascular, and other tissue and organ lesions. Implementing preventative measures is crucial to prevent the onset of diabetes. This paper proposes a novel integrated method based on FT-Transformer for effectively identifying diabetes-related drug active compounds. The proposed method integrates data features using FT-Transformer while incorporating local interactions through Random Forest. This approach dynamically and linearly weights the predictive probabilities from both the tree model and Transformer, overcoming limitations of conventional integration methods that rely on a single model type. The experimental data were obtained from the latest diabetes-related literature. Results demonstrate that our proposed integration method outperforms traditional classifiers across multiple metrics including AUC, Sensitivity, Specificity, Kappa, MCC, F1 score, and PR curves in identifying diabetes-related compounds.