M. Faris Al Hakim, Budi Prasetiyo, Jumanto, Much Aziz Muslim
Convolutional Neural Network (CNN) becomes the robust deep learning model for image classification. It has many configurations applied to achieve the best performance. A high number of data is needed to build a good model of CNN. It makes small data cannot support optimally to build a robust model. In fact, certain cases have limited sources of data. This study aimed to build CNN robust model using small dataset. In this research, several models were proposed and vehicle miniature dataset was selected. Configurations for optimizer, input size, batch size, and pooling were also done to obtain the expected model. Data augmentation became an experiment for comparing the model performance. Based on proposed model, the results show that model obtain 84.87% of accuracy using data augmentation and more than equal to 98.69% of accuracy without data augmentation. Moreover, this study is also able to increase accuracy score when compared to previous research. © 2023 IEEE.
Universitas Negeri Semarang, Department of Computer Science, Semarang, Indonesia; Universiti Tun Hussein Onn Malaysia, Faculty of Technology Management & Bussiness, Johor, Malaysia
