The Impact of Data Augmentation on the Accuracy of Convolutional Neural Network Models for Skin Cancer Classification Using the Skin Cancer MNIST HAM10000 Dataset

Pramesti, Nadia (2025) The Impact of Data Augmentation on the Accuracy of Convolutional Neural Network Models for Skin Cancer Classification Using the Skin Cancer MNIST HAM10000 Dataset. Tugas Akhir thesis, University of Technology Yogyakarta.

Text #Abstrak_Nadia Pramesti_5211811010_Sains Data.pdf Download (174kB)

Abstract

ABSTRACT Skin cancer is a disease that can be identified through dermatoscopic images. This study aims to evaluate the impact of data augmentation on the accuracy of a Convolutional Neural Network (CNN) model in classifying seven types of skin cancer using the HAM10000 dataset. The ResNet50 architecture was employed as the primary model through a transfer learning approach. Three experimental scenarios were conducted: a baseline model without augmentation, Type 1 augmentation (rotation, zoom, and translation), and Type 2 (mild augmentation). To address class imbalance, resampling was performed to ensure each class contained 500 samples. The baseline model achieved an accuracy of 75.48%, though it demonstrated poor performance on minority classes. Type 1 augmentation yielded 71.31% accuracy with improved performance uniformity across classes. Type 2 augmentation resulted in the highest accuracy of 77.37%, along with stable performance across all categories. These findings suggest that data augmentation, particularly mild augmentation techniques, can enhance the generalization capability of CNNs and contribute to more balanced skin cancer classification outcomes. Keywords: Classification, Skin Cancer, CNN, Data Augmentation, ResNet50, HAM10000.

Item Type:	Thesis (Skripsi, Tugas Akhir or Kerja Praktek) (Tugas Akhir)
Subjects:	T Technology > T Technology (General)
Divisions:	Fakultas Sains Dan Teknologi > Data Science
Depositing User:	Sains Data
Date Deposited:	08 Aug 2025 02:09
Last Modified:	08 Aug 2025 02:09
URI:	http://eprints.uty.ac.id/id/eprint/18412

Actions (login required)

View Item