Quantitative Analysis of Major Compounds in Coffee Using Raman Spectroscopy Technique Combined with Data Analysis and Machine Learning

Abstract

This article presents a method for analyzing caffeine, which is a primary compound in coffee, by quantitatively analyzing standard chemical substances using Raman Spectroscopy technique together with data analysis using Machine Learning. In this experiment, standard caffeine samples were prepared at concentrations of 500 ppm, 1000 ppm, 1500 ppm, 2000 ppm, 3000 ppm, and 4000 ppm. Subsequently, these samples were measured using Raman Spectroscopy technique to obtain a set of Raman spectra data. However, the quality of the Raman spectra data was found to be insufficient, necessitating enhancement of the data quality through preprocessing techniques such as Baseline Removal, Cosmic ray Removal, Smoothing Signal, Outlier Removal, Amplitude  and Normalization Following this, the dataset containing 80% of the total data (Training Dataset), was used to train model, while the remaining 20% (Validation Dataset) was used for testing and parameter adjustment to improve the model’s prediction efficiency for caffeine quantitates. This research compared the performance of all 18 models using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Multiple Determination (R-Square) as performance indicators. The model with highest performance according to these indicators was the Extra Tree Regressor, with MAE, RMSE, and R-square values of 710 ppm, 914,05 ppm, and 0.34 respectively. The methods and experimental results from this study can be further developed for determining caffeine quantities in coffee.