Klasifikasi Animals Dataset Menggunakan Metode Ensemble CNN

Classification animal datasets based on images is an interesting challenge in the world of image processing and machine learning. In this research, we propose an effective classification method for identifying images in animal datasets using the Ensemble Convolutional Neural Network (CNN) approach. This research compares the performance of the proposed method model EfficientNet-B2 and the use of ensemble methods in detecting animal types. The dataset used is an image dataset of 10 types of animals. The number of datasets taken was 26,179. 20943 for training, 2095 for validation, and 3141 for testing. The model was built by implementing transfer learning and fine tuning five times training using the same architecture and parameters. Model optimization was then carried out by applying the ensemble voting method to the five models that had been trained. Based on the research results, the accuracy values of the five models are 97.90%, 97.85%, 97.99%, 98.28%, and 98.00%, respectively. After applying the ensemble method by voting on the classification prediction results for each model, an accuracy result of 98.52% was obtained. Based on the results, the accuracy of this research is higher than in several previous studies and it can be concluded that the use of the ensemble voting method in the proposed method model EfficientNet-B2 is successful and capable of classifying dataset images.