Small chemicals that block the human ether-a-go-go-related (hERG) potassium channel result in a prolonged QT interval, which can have serious cardiotoxic effects and is a major factor in drug development failures. In order to develop the drug successfully, molecular image classification prediction of hERG blockers is essential for designing drug candidates without the risk of cardiotoxicity. Convolutional neural networks (CNNs) that were pre-trained, frozen, and fine-tuned using data augmentation, along with Imagenet and VGG19, were used to construct models through transfer learning. According to the findings, the models’ predictive power and resilience can be greatly increased by utilizing data augmentation and transfer learning. The VGG19 transfer learning model showed the test dataset’s accuracy values 97% following data augmentation, and the after using frozen pre-trained CNN, the accuracy of the test dataset was 99%. Model performance was enhanced with transfer learning approaches and significantly predicted cardiotoxicity of the compounds. Finally, we conclude that image classification model, in combination with transfer learning, offers a potentially novel approach for predicting the cardiotoxicity of drug candidates.