Work

Improving Image Classification Accuracy With Oversampling and Data Augmentation Using Deep Learning: a Case Study on the Simpsons Characters Dataset

Year: 2024

Type: article

Abstract: The issue of data imbalance in image classification often hinders deep learning models from making accurate predictions, especially for minority classes. This study introduces AugOS-CNN (Augmentation and Over Sampling with CNN), a novel approach that combines oversampling and data augmentation techniques to address data imbalance. The The Simpsons Characters dataset is used in this study, featuring five main character classes: Bart, Homer, Agnes, Carl, and Apu. The number of samples in each class is balanced to 2,067 using an augmentation method based on Augmentor. The proposed model integrates oversampling and augmentation steps with a Convolutional Neural Network (CNN) architecture to improve classification accuracy. Evaluation results show that the AugOS-CNN model achieves the highest accuracy of 96%, outperforming the baseline CNN approach without data balancing techniques, which only reaches 91%. These findings demonstrate that the AugOS-CNN model effectively enhances image classification performance on datasets with imbalanced class distributions, contributing to the development of more robust deep learning methods for addressing data imbalance issues.

Source: Jurnal Riset Informatika

DOI: 10.34288/jri.v6i4.348

Authors: Ilham Maulana Ilham, Siti Ernawati, Muhammad Indra

Institution: Universitas Nusa Mandiri

Language: English

Topic: Imbalanced Data Classification Techniques

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

SDG: Reduced inequalities

Open Access status: gold

Estimated APC paid: $32