OPTIMIZATION OF CONVOLUTIONAL RECURRENT NEURAL NETWORK (CRNN) MODEL WITH DATA AUGMENTATION FOR CIFAR-10 IMAGE CLASSIFICATION

Abstrak

Image classification is a crucial component in the advancement of arti-ficial intelligence (AI) technology, which has been widely applied across various sectors, such as facial recognition systems and medical diagno-sis. One of the most commonly used benchmark datasets for evaluating classification model performance is CIFAR-10, which consists of 60,000 color images sized 32×32 pixels, divided into 10 categories. However, the relatively small size of the dataset makes overfitting a major challenge in achieving optimal model generalization. In this study, the Convolutional Recurrent Neural Network (CRNN) model is optimized using data augmentation techniques—such as horizontal flipping, random cropping, rotation, and noise addition—to enrich the training data diversity and reduce overfitting. The evaluation results show a 3.5% increase in accuracy compared to models trained without augmentation, along with improvements in precision, recall, and F1-score, indicating a more balanced performance in detection and classi-fication. This study also opens opportunities for further development by applying more advanced augmentation techniques, exploring deeper model architectures, incorporating additional regularization methods, and integrating synthetic data through Generative Adversarial Networks (GANs) and transfer learning strategies to enhance the model’s efficien-cy and adaptability. The findings from this research are expected to provide practical and methodological contributions toward building robust, efficient, and more adaptive deep learning models capable of addressing the challenges of complex image data in the future.

Identifier

• OpenAlex: https://openalex.org/W7166571475
• DOI: https://doi.org/10.29100/jipi.v11i2.8230