Electronic Theses and Dissertation

Penelitian ini bertujuan mengevaluasi kinerja model Support Vector Machine (SVM) dalam mengklasifikasikan minyak kenanga (Cananga odorata) berdasarkan varietas dan asal geografis menggunakan data Fourier Transform Infrared (FTIR), dengan Partial Least Squares-Discriminant Analysis (PLS-DA) sebagai metode pembanding. Sampel penelitian terdiri atas minyak Cananga odorata forma macrophylla dan Cananga odorata var. fruticosa yang berasal dari Aceh Selatan dan Bireuen. Sebanyak 40 spektrum FTIR digunakan sebagai data latih, sedangkan 10 spektrum minyak kenanga asal Singkil digunakan sebagai data uji eksternal. Data dipraproses menggunakan Standard Normal Variate (SNV), turunan pertama Savitzky–Golay, dan standardisasi. Model SVM pada Rezim A menghasilkan nilai akurasi, balanced accuracy, F1-score, MCC, Cohen's Kappa, dan AUC masing-masing sebesar 1,000. Namun, pada evaluasi CROSS dan Leave-One-Group-Out (LOGO), nilai balanced accuracy menurun menjadi 0,500 dan 0,000 akibat pseudo-replikasi. Pada klasifikasi varietas, SVM memberikan balanced accuracy sebesar 0,500, lebih tinggi dibandingkan PLS-DA sebesar 0,450, sedangkan pada klasifikasi asal geografis kedua metode menghasilkan balanced accuracy yang sama, yaitu 0,500. Validasi eksternal berhasil mengidentifikasi sampel Singkil sebagai Cananga odorata var. fruticosa dengan akurasi 100% dan memprediksi asal geografis terdekat di Bireuen dengan tingkat keyakinan 0,66. Hasil penelitian menunjukkan bahwa kombinasi FTIR dan SVM berpotensi diterapkan sebagai metode autentikasi minyak kenanga yang cepat, non-destruktif, dan efektif.

Kata kunci: minyak kenanga, FTIR, Support Vector Machine (SVM), PLS-DA, kemometrik.

This study aimed to evaluate the performance of the Support Vector Machine (SVM) model for classifying ylang-ylang (Cananga odorata) essential oil based on variety and geographical origin using Fourier Transform Infrared (FTIR) data, with Partial Least Squares-Discriminant Analysis (PLS-DA) employed as a comparative method. The samples consisted of Cananga odorata forma macrophylla and Cananga odorata var. fruticosa collected from Aceh Selatan and Bireuen. A total of 40 FTIR spectra were used as the training dataset, while 10 FTIR spectra from a Singkil sample were used for external validation. Spectral data were preprocessed using Standard Normal Variate (SNV), the first derivative of Savitzky–Golay, and standardization. Under Regime A, the SVM model achieved accuracy, balanced accuracy, F1-score, Matthews Correlation Coefficient (MCC), Cohen's Kappa, and Area Under the Curve (AUC) values of 1.000. However, balanced accuracy decreased to 0.500 and 0.000 under CROSS and Leave-One-Group-Out (LOGO) evaluations, respectively, due to pseudo-replication. For variety classification, SVM achieved a balanced accuracy of 0.500, exceeding the PLS-DA value of 0.450, whereas both methods produced an identical balanced accuracy of 0.500 for geographical origin classification. External validation correctly identified the Singkil sample as Cananga odorata var. fruticosa with 100% accuracy and predicted Bireuen as the closest geographical origin with a confidence score of 0.66. These findings demonstrate that the combination of FTIR and SVM has strong potential as a rapid, non-destructive, and effective approach for authenticating ylang-ylang essential oil according to variety and geographical origin. Keywords: cananga oil, FTIR, Support Vector Machine (SVM), PLS-DA, chemometrics.