The apple industry suffers considerable economic losses each year as a result of diseases and pests, emphasising the critical need for precise and quick detection procedures. The paper carefully examines the use of ensemble techniques to increase the accuracy of apple disease detection. The research begins with the creation of a comprehensive dataset that includes images of healthy and diseased apples, as well as extensive data collection and labeling. To solve the challenge of distinguishing various diseases with similar symptoms, the study employs an ensemble technique, which draws on the strengths of many models. The ensemble incorporates a Convolutional Neural Network (CNN) model trained via transfer learning to combine several characteristics taken from raw images. To reduce overfitting, data augmentation methods including rotation, zooming, and scaling are deliberately used. This research not only provides a useful tool for farmers, but it also helps to reduce agricultural losses and promote economic growth by allowing for exact and complete identification of apple diseases. The research compares the performance of a single pre-trained model and several ensemble learning models in categorizing apple disease images. The effectiveness of the proposed ensemble model is evaluated using a publicly accessible dataset of four apple fruit classifications. DenseNet121, a single pre-trained model, achieves 90.05% accuracy. Furthermore, a weighted average ensemble of three models (VGG16, DenseNet121, and InceptionV3) outperforms other ensemble models, with an accuracy of 94.24%.