ISNENI, SYILMI (2026) OPTIMALISASI KLASIFIKASI CITRA DAUN TANAMAN HERBAL UNTUK PENYAKIT KULIT MENGGUNAKAN RESNET, PRINCIPAL COMPONENT ANALYSIS, DAN LOGISTIC REGRESSION. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
01 COVER.pdf Download (471kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
02 BAB 1.pdf Restricted to Registered users only Download (50kB) |
|
|
Text (BAB II)
03 BAB 2.pdf Restricted to Registered users only Download (61kB) |
|
|
Text (BAB III)
04 BAB 3.pdf Restricted to Registered users only Download (232kB) |
|
|
Text (BAB IV)
05 BAB 4.pdf Restricted to Registered users only Download (262kB) |
|
|
Text (BAB V)
06 BAB 5.pdf Restricted to Registered users only Download (35kB) |
|
|
Text (DAFTAR PUSTAKA)
07 DAFTAR PUSTAKA.pdf Restricted to Registered users only Download (89kB) |
|
|
Text (LAMPIRAN)
08 LAMPIRAN.pdf Restricted to Registered users only Download (514kB) |
Abstract
Skin diseases are among the most prevalent health issues in Indonesian society, and various local herbal plants are known to have potential as natural treatment alternatives. However, the manual identification of herbal plants requires specialized expertise and is prone to errors due to high visual similarity between species. This study aims to optimize a classification system for herbal plant leaf images related to skin diseases using a three-stage pipeline: deep feature extraction with the Residual Network (ResNet-50) architecture, dimensionality reduction with Principal Component Analysis (PCA), and final classification using Logistic Regression. ResNet-50 is employed to extract rich and discriminative feature representations from herbal plant leaf images. PCA is applied to reduce the dimensionality of extracted feature vectors from 2,048 to 128 dimensions without losing essential information, thereby significantly reducing computational load. Logistic Regression is selected as the final classifier for its efficiency on low-dimensional PCA features, its ability to produce prediction probabilities, and its resistance to overfitting through L2 regularization. The dataset used is the Indonesian Herb Leaf Dataset 3500, which consists of 3,500 leaf images from 10 classes of Indonesian herbal plants, with 350 images per class. The results show that the ResNet-50 + PCA + Logistic Regression pipeline achieves 98.29% accuracy, 98.32% precision (macro), 98.29% recall (macro), and 98.28% F1-Score macro on the test data, with 99.94% ROC-AUC. Five-fold cross-validation yields a mean accuracy of 96.48% with a standard deviation of +-0.99%, indicating stable performance. Comparison with other experiments proves that this pipeline significantly outperforms PCA Only (64.86%) and LR Only (66.86%), and is only slightly lower than ResNet Only (99.43%) with the advantage of better computational efficiency. The resulting system is designed to support the herbal plant classification module within the HerbaScan platform. Keywords: Image Classification, Herbal Plants, Skin Diseases, ResNet-50, Principal Component Analysis (PCA), Logistic Regression, Deep Learning, HerbaScan. Penyakit kulit merupakan salah satu masalah kesehatan yang banyak ditemui di masyarakat Indonesia, dan berbagai jenis tanaman herbal lokal diketahui memiliki potensi sebagai alternatif pengobatan alami. Namun, proses identifikasi tanaman herbal secara manual membutuhkan keahlian khusus dan rentan terhadap kesalahan akibat kemiripan visual antar spesies. Penelitian ini bertujuan untuk mengoptimalkan sistem klasifikasi citra daun tanaman herbal yang berkaitan dengan penyakit kulit menggunakan pipeline tiga tahap: ekstraksi fitur mendalam dengan arsitektur Residual Network (ResNet-50), reduksi dimensi dengan Principal Component Analysis (PCA), serta klasifikasi akhir menggunakan Logistic Regression. ResNet-50 digunakan untuk mengekstraksi representasi fitur yang kaya dan diskriminatif dari citra daun tanaman herbal. PCA diterapkan untuk menyederhanakan dimensi vektor fitur hasil ekstraksi dari 2.048 menjadi 128 dimensi tanpa menghilangkan informasi esensial, sehingga beban komputasi berkurang secara signifikan. Logistic Regression dipilih sebagai classifier akhir karena efisiensinya pada fitur berdimensi rendah hasil PCA, kemampuannya menghasilkan probabilitas prediksi, serta ketahanannya terhadap overfitting melalui regularisasi L2. Dataset yang digunakan adalah Indonesian Herb Leaf Dataset 3500 yang terdiri atas 3.500 citra dari 10 kelas tanaman herbal Indonesia, dengan masing-masing kelas terdiri dari 350 citra. Hasil penelitian menunjukkan bahwa pipeline ResNet-50 + PCA + Logistic Regression mencapai akurasi 98,29%, precision (macro) 98,32%, recall (macro) 98,29%, dan F1-Score macro 98,28% pada data uji, serta ROC-AUC 99,94%. Validasi silang 5-fold menghasilkan mean akurasi 96,48% dengan standar deviasi +-0,99%, menunjukkan performa yang stabil. Perbandingan dengan eksperimen lain membuktikan bahwa pipeline ini jauh mengungguli PCA Only (64,86%) dan LR Only (66,86%), serta hanya berbeda tipis dari ResNet Only (99,43%) dengan keunggulan efisiensi komputasi yang lebih baik. Sistem yang dihasilkan dirancang untuk mendukung modul klasifikasi tanaman herbal dalam platform HerbaScan. Kata kunci: Klasifikasi Citra, Tanaman Herbal, Penyakit Kulit, ResNet-50, Principal Component Analysis (PCA), Logistic Regression, Deep Learning, HerbaScan.
Actions (login required)
 |
View Item |
