Electronic Theses and Dissertation

Minyak kayu putih merupakan produk alami yang banyak dimanfaatkan dalam bidang kesehatan, kosmetik, dan farmasi. Variasi mutu produk termasuk keberadaan minyak yang teradulterasi menimbulkan tantangan dalam memastikan kualitas dan kemurniannya. Penelitian ini bertujuan untuk mengembangkan metode klasifikasi minyak kayu putih menggunakan pendekatan Machine Learning berbasis data spektrum Fourier Transform Infrared (FTIR). Sampel yang diuji terdiri atas minyak kayu putih murni hasil penyulingan rakyat, minyak kayu putih teradulterasi, dan minyak kayu putih komersial. Analisis data spektrum dilakukan melalui reduksi dimensi menggunakan Principal Component Analysis (PCA) dengan penerapan lima teknik preprocessing yaitu 2nd derivative, smoothing, SNV, MSC, dan baseline correction untuk meningkatkan kualitas data serta klasifikasi menggunakan Support Vector Machine (SVM). Hasil penelitian menunjukkan bahwa spektrum FTIR dapat mengidentifikasi gugus fungsi khas dari senyawa 1,8-cineole, serta indikasi adanya α-pinene dalam sampel teradulterasi. Preprocessing berhasil meningkatkan efektivitas PCA dengan teknik baseline correction yang meningkatkan cumulative variance dari 96% menjadi 99%. PCA juga dapat mendeteksi threshold penambahan terpentin hingga 4,5%. Evaluasi klasifikasi SVM menunjukkan bahwa rentang fingerprint (1500–400 cm⁻¹) menghasilkan akurasi tertinggi, dengan nilai training dan validation accuracy sebesar 94,54% dan 94,55%. Kombinasi preprocessing, PCA, dan SVM menghasilkan akurasi optimal dengan nilai training accuracy sebsesar 95,83% dan validation accuracy sebesar 94,44%. Pendekatan ini efektif diterapkan untuk klasifikasi minyak kayu putih dan memiliki potensi besar untuk standardisasi mutu produk minyak atsiri.

Kata Kunci: Minyak Kayu Putih, FTIR, Preprocessing, PCA, SVM

Cajuput oil is a product widely utilized in the fields of health, cosmetics, and pharmaceuticals. Variations in product quality, including the presence of adulterated oil, pose challenges in ensuring its authenticity and purity. This study aims to develop a classification method for cajuput oil using a Machine Learning approach based on Fourier Transform Infrared (FTIR) spectrum data. The samples analyzed consist of pure cajuput oil from traditional distillation, adulterated cajuput oil, and commercial cajuput oil. Spectrum data analysis was carried out through dimensionality reduction using Principal Component Analysis (PCA) with the application of five preprocessing techniques, namely 2nd derivative, smoothing, Standard Normal Variate (SNV), Multiple Scatter Correction (MSC), and baseline correction to enhance data quality, followed by classification using Support Vector Machine (SVM). The results indicate that the FTIR spectrum can identify characteristic functional groups of the compound 1,8-cineole, as well as the presence of α-pinene in the adulterated samples. Preprocessing successfully enhanced the effectiveness of PCA with the baseline correction technique, which increased the cumulative variance from 96% to 99%. PCA also identified the threshold of turpentine addition up to 4.5%. SVM classification evaluation revealed that the fingerprint region (1500–400 cm⁻¹) yielded the highest accuracy, with training and validation accuracy values of 94.54% and 94.55%, respectively. The combination of preprocessing, PCA, and SVM produced optimal classification performance, achieving 95.83% training accuracy and 94.44% validation accuracy. This approach is effectively applied for the classification of cajuput oil and has great potential for the standardization of essential oil product quality. Keyword: Cajuput Oil, FTIR, Preprocessing, PCA, SVM