Please use this identifier to cite or link to this item: https://repositori.uma.ac.id/handle/123456789/27363
Title: Analisis performa convolution neural network untuk klasifikasi hewan berdasarkan perbedaan ukuran kernels
Other Titles: Performance analysis of convolution neural networks for animal classification based on different kernel sizes
Authors: Pane, Ilham M
metadata.dc.contributor.advisor: Sembiring, Arnes
Keywords: Convolutional Neural Network;image classification;kernel size;confusion matrix;accuracy;klasifikasi citra;ukuran kernel;akurasi
Issue Date: May-2025
Publisher: Universitas Medan Area
Series/Report no.: NPM;198160079
Abstract: Adapun tujuannya dari penelitian ini guna menganalisis pengaruh variasi ukuran kernel pada arsitektur Convolutional Neural Network (CNN) terhadap performa klasifikasi citra hewan. Ukuran kernel yang diuji meliputi 3x3, 5x5, 7x7, dan 9x9. Evaluasi dilakukan dengan metrik akurasi maupun analisis confusion matrix guna mengukur efektivitas setiap model. Hasil menunjukkan bahwasanya kernel berukuran 5x5 memberikan akurasi tertinggi dan distribusi klasifikasi yang paling seimbang, sedangkan kernel 9x9 menghasilkan penurunan performa yang signifikan. Ukuran kernel yang terlalu besar menyebabkan model kehilangan kemampuan dalam menangkap fitur lokal, sehingga terjadi kesalahan klasifikasi yang tinggi. Sebaliknya, ukuran kernel yang moderat mampu menjaga keseimbangan antara cakupan informasi global dan detail lokal. Temuan ini menegaskan pentingnya pemilihan ukuran kernel yang tepat dalam perancangan CNN untuk mencapai hasil klasifikasi yang optimal. This study aims to analyze the impact of kernel size variation in Convolutional Neural Network (CNN) architectures on the performance of animal image classification. The kernel sizes evaluated include 3x3, 5x5, 7x7, and 9x9. Performance was assessed using accuracy metrics and confusion matrix analysis to determine the effectiveness of each model. The results indicate that the 5x5 kernel achieved the highest accuracy and the most balanced classification distribution, while the 9x9 kernel resulted in a significant decline in performance. Excessively large kernels led to the mode’sl inability to capture local features, causing a high rate of misclassification. In contrast, moderately sized kernels maintained a balance between capturing global context and preserving local detail. These findings highlight the importance of selecting an appropriate kernel size in CNN architecture design to achieve optimal classification results
Description: 14 Halaman
URI: https://repositori.uma.ac.id/handle/123456789/27363
Appears in Collections:SP - Informatic Engineering

Files in This Item:
File Description SizeFormat 
198160079 - Ilham M Pane - Fulltext.pdfFulltext877.71 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.