In recent years, various focusing on Dipeptidyl Peptidase-4 inhibitors drugs discovery to achieve better treatments for type II Diabetes Mellitus. As such, new medical research on new DPP-4 inhibitors with minimal effects is still crucial. One of the drug designs based on in silico is a virtual screening-based ligand (LBVS). The LBVS method used in this research is Quantitative structure-activity relation (QSAR). The QSAR model is a fast and cost-effective alternative for experimental measurement in drug discovery. Deep learning has also been successful and is now widely used in drug discovery. In this study, we propose a combination of two deep learning approaches, namely the Conv1D-LSTM model as a renewable method for predicting the classification of Dipeptidyl Peptidase-4 inhibitors. This model includes the Conv1D model as a data encoding stage and LSTM as a model for the classification of compounds in Dipeptidyl Peptidase-4 inhibitors. We use 2604 molecular structures of DPP-4 inhibitors with 1443 active compounds and 1161 inactive compounds. The result in our proposed model has great accuracy for the classification of compounds in the Dipeptidyl Peptidase-4 inhibitors with an accuracy of 86.18%. Furthermore, the values for sensitivity, specificity, and MCC were obtained are 91.05%, 79.45%, and 71.50% respectively.