Cassava is a crop that has high demand in Indonesia, marked by increasing production levels over time. In addition to quantity, crop quality must be maintained, one of which is by paying attention to disease symptoms. Disease symptoms that appear on cassava leaves can be detected by visual inspection. However, more knowledge is needed to distinguish the symptoms of one disease from another. One solution to this problem is the use of convolutional neural networks (CNN) for disease classification. The author uses a CNN model for this problem. The performance assessment parameters of the CNN model used are accuracy, precision, recall, and F1-score. This study will use two architectures in transfer learning, namely EfficientNet-B4 and Inception-V3. Both of these architectures are still rarely used in related case studies. The purpose of increasing the number of parameters is to find the optimal configuration of the optimizer and learning rate that can maximize model performance. By increasing the number of parameters and utilizing two architectures in transfer learning, it is hoped that the model's ability to handle the complexity of the problem of classifying images of cassava leaves with disease can be improved. The focus of this study will also be focused on the application of the EfficientNet-B4 and Inception-V3 architectures with a hyperparameter tuning scheme to improve model performance. Therefore, this research is expected to provide a superior contribution in the development of CNN for disease classification in cassava leaves, with better and more accurate performance.