Optimasi Sifat Mekanik Edible Film Menggunakan Algoritma Machine Learning

Abstract

Pengembangan edible film sebagai kemasan pangan ramah lingkungan menghadapi tantangan dalam memenuhi sifat mekanik yang optimal karena prediksi kuat tarik dan elongasi secara konvensional masih memakan waktu dan biaya tinggi. Penelitian ini bertujuan menganalisis performa algoritma machine learning dalam memprediksi sifat mekanik edible film, mengidentifikasi variabel paling berpengaruh, serta menentukan kombinasi bahan optimal. Metode penelitian meliputi pemodelan prediktif menggunakan Random Forest, SVR, dan XGBoost terhadap data eksperimental sifat mekanik edible film dengan empat skema perbandingan data (90:10, 80:20, 70:30, dan 60:40). Kinerja model dievaluasi menggunakan koefisien determinasi (R²) dan Mean Squared Error, serta dioptimasi melalui penyesuaian hiperparameter dengan pendekatan multi-objective. Hasil penelitian menunjukkan bahwa XGBoost dengan rasio 60:40 memberikan kinerja terbaik, gliserol merupakan variabel yang paling berpengaruh terhadap kuat tarik dan elongasi, serta optimasi menghasilkan kombinasi bahan terbaik yang memberikan keseimbangan optimal antara kekuatan tarik dan fleksibilitas.

The development of edible films as environmentally friendly food packaging faces challenges in achieving optimal mechanical properties due to the time-consuming and high-cost nature of conventional methods for predicting tensile strength and elongation. This study aims to analyze the performance of machine learning algorithms in predicting the mechanical properties of edible films, identify the most influential variables, and determine the optimal material combination. The research methodology involves predictive modeling using Random Forest, SVR, and XGBoost based on experimental mechanical property data, employing four data-splitting schemes (90:10, 80:20, 70:30, and 60:40). Model performance was evaluated using the coefficient of determination (R²) and Mean Squared Error, followed by optimization through hyperparameter tuning with a multi-objective approach. The results indicate that XGBoost with a 60:40 data ratio achieves the best performance, glycerol is the most influential variable affecting tensile strength and elongation, and the optimization process yields an optimal material combination that provides a balanced improvement in strength and flexibility.