Improved prediction of drug-drug interactions using ensemble deep neural networks

•A novel model for predicting drug-drug interactions with high performance.•Different state-of-the-art features have been calculated from chemical information and used in the models.•Ensemble model is created by stacking three optimal models: Random Forest, XGBoost, and deep neural network.•Compared with previous predictors, our model had a significant improvement in training and testing processes.•A useful resource to facilitate the detection of drug-drug interactions to support medical decisions and drug development.