Pemodelan Data Rotational Anisotropy Second Harmonic Generation pada Malachite Green Menggunakan Machine Learning

Abstract

Malachite Green (MG) merupakan kontaminan pangan toksik yang penting untuk dideteksi pada permukaan. Teknik Second Harmonic Generation (SHG) sensitif terhadap molekul teradsorpsi, namun pemodelan polanya umumnya bergantung pada model fisis yang memerlukan asumsi tertentu. Penelitian ini menerapkan dan membandingkan pendekatan machine learning berbasis data untuk memodelkan pola intensitas SHG MG pada permukaan silikon sebagai fungsi sudut polarisasi, serta menilai posisinya terhadap Simplified Bond Hyperpolarizability Model (SBHM). Sudut polarisasi ditransformasi menjadi fitur sinus dan kosinus, lalu Support Vector Regression, Polynomial Regression, dan Random Forest dilatih pada data eksperimen tujuh variasi konsentrasi. Support Vector Regression dan Polynomial Regression derajat enam menunjukkan performa terbaik, sedangkan Random Forest kurang sesuai untuk pola kontinu RA-SHG. Kedua model merekonstruksi kurva setara atau lebih baik daripada SBHM secara numerik, tetapi diposisikan sebagai pelengkap karena SBHM unggul dalam interpretasi fisis.

Malachite Green (MG) is a toxic food contaminant that is important to detect on surfaces. Second Harmonic Generation (SHG) is highly sensitive to adsorbed molecules; however, its pattern modeling generally relies on physical models that require specific assumptions. This study applies and compares data-driven machine learning approaches to model the SHG intensity pattern of MG on a silicon surface as a function of polarization angle and evaluates their relationship to the Simplified Bond Hyperpolarizability Model (SBHM). The polarization angle was transformed into sine and cosine features, and Support Vector Regression, Polynomial Regression, and Random Forest were Trained using Experimental data from seven concentration variations. Support Vector Regression and sixth-degree Polynomial Regression showed the best performance, while Random Forest was less suitable for the continuous RA-SHG pattern. Both models reconstructed the curves with numerical performance comparable to or better than SBHM, although they are positioned as complementary approaches because SBHM remains superior in providing physical interpretation.