Prediksi Kandungan Proksimat Biji Kopi Robusta Menggunakan Near Infrared Spectroscopy dengan Metode PLS dan MLR

Abstract

Kopi Robusta (Coffea canephora) merupakan salah satu komoditas perkebunan unggulan Indonesia. Penentuan kandungan proksimat biji kopi yang meliputi kadar air, protein, lemak, kadar abu, dan karbohidrat selama ini masih dilakukan secara destruktif di laboratorium yang membutuhkan waktu lama dan biaya tinggi. Near Infrared Spectroscopy (NIRS) dikembangkan sebagai metode alternatif yang lebih efisien karena bersifat tidak merusak (nondestructive), cepat, dan akurat. Penelitian ini bertujuan memprediksi kandungan proksimat biji kopi robusta secara nondestruktif menggunakan NIRS dengan metode Partial Least Squares (PLS) dan Multiple Linear Regression (MLR). Sebanyak 60 sampel biji kopi robusta (60 g) diukur reflektansinya dengan alat spektrometer NIRFlex N-500 pada panjang gelombang 1000-2500 nm. Pre-treatment yang digunakan yaitu Normalization (No1), Standard Normal Variate (SNV), Multiplicative Scatter Correction (MSC), Savitzky-Golay 1st Derivative (SG1), kombinasi No1SG1, dan SNVSG1. Prediksi terbaik untuk kadar air diperoleh dengan PLS pre-treatment SNV faktor 6 (r = 0,78; RPD = 1,61; konsistensi = 109,77%). Prediksi terbaik protein diperoleh dengan PLS data Original faktor 8 (r = 0,74; RPD = 1,51; konsistensi = 100,74%). Prediksi terbaik lemak diperoleh dengan PLS pre-treatment SG1 faktor 5 (r = 0,78; RPD = 1,20; konsistensi = 82,83%). Prediksi terbaik kadar abu diperoleh dengan PLS pre-treatment Normalisasi faktor 8 (r = 0,70; RPD = 1,34; konsistensi = 91,89%). Prediksi terbaik karbohidrat diperoleh dengan PLS pre-treatment SNV faktor 7 (r = 0,82; RPD = 1,78; konsistensi = 110,01%). Prediksi kadar air, karbohidrat, dan protein dengan NIRS dapat digunakan untuk skrining awal (RPD = 1,5), sedangkan prediksi lemak dan kadar abu belum dapat diandalkan karena nilai RPD di bawah 1,5. Metode PLS menghasilkan keakuratan lebih tinggi dibandingkan MLR pada semua parameter.

Robusta coffee (Coffea canephora) is one of Indonesia's leading plantation commodities. The determination of proximate composition of coffee beans, which includes moisture content, protein, fat, ash content, and carbohydrate, is currently performed destructively in the laboratory, requiring substantial time and cost. Near Infrared Spectroscopy (NIRS) has been developed as a more efficient alternative method due to its nondestructive, rapid, and accurate nature. This study aimed to predict the proximate composition of Robusta coffee beans nondestructively using NIRS with the Partial Least Squares (PLS) and Multiple Linear Regression (MLR) methods. A total of 60 Robusta coffee bean samples (60 g) were measured for their reflectance using a NIRFlex N-500 spectrometer at a wavelength range of 1000-2500 nm. The pre-treatments applied were Normalization (No1), Standard Normal Variate (SNV), Multiplicative Scatter Correction (MSC), Savitzky-Golay 1st Derivative (SG1), and combinations of No1SG1 and SNVSG1. The best prediction for moisture content was obtained using PLS with SNV pre-treatment at factor 6 (r = 0.78; RPD = 1.61; consistency = 109.77%). The best prediction for protein was obtained using PLS with Original data at factor 8 (r = 0.74; RPD = 1.51; consistency = 100.74%). The best prediction for fat was obtained using PLS with SG1 pre-treatment at factor 5 (r = 0.78; RPD = 1.20; consistency = 82.83%). The best prediction for ash content was obtained using PLS with Normalization pre-treatment at factor 8 (r = 0.70; RPD = 1.34; consistency = 91.89%). The best prediction for carbohydrate was obtained using PLS with SNV pre-treatment at factor 7 (r = 0.82; RPD = 1.78; consistency = 110.01%). NIRS prediction of moisture content, carbohydrate, and protein can be used for preliminary screening (RPD = 1.5), while fat and ash content predictions are not yet reliable because the RPD values are below 1.5. The PLS method produced higher accuracy than MLR for all parameters.