Related Experiment Video
Updated: Mar 2, 2026
10:13
Using Capillary Electrophoresis to Quantify Organic Acids from Plant Tissue: A Test Case Examining Coffea arabica Seeds
Published on: November 12, 2016
10.3K
Prediction of SCA Scores in Specialty Coffee Using Machine Learning
Gabriel Rezende Ferraz1, Felipe André Oliveira Freitas1, Harim H Baldi1
1Universidade de Sao Paulo (USP), Piracicaba, Brasil.
Journal of Food Science
|February 28, 2026
Summary
Machine learning models can predict Specialty Coffee Association (SCA) scores using coffee processing data. Variable selection offers a robust and efficient method for estimating coffee quality, reducing reliance on extensive sensory evaluations.
Area of Science:
- Agricultural Science
- Data Science
- Food Science
Background:
- Specialty coffee quality is crucial for its global commodity value.
- Specialty Coffee Association (SCA) scores are key commercial indicators.
- Sensory evaluations for SCA scores are resource-intensive and variable.
Purpose of the Study:
- To develop predictive models for estimating SCA scores.
- To utilize processing and production variables for quality prediction.
- To compare different modeling approaches for efficiency and robustness.
Main Methods:
- Collected processing data (2019-2023) covering reception to storage.
- Applied Random Forest (RF) and XGBoost (XGB) regression algorithms.
- Evaluated three approaches: complete variables, PCA, and selected variables.
Main Results:
- The RF model with all variables showed the best performance (MAE=0.80, RMSE=1.03, R²=0.53).
- Models with seven selected variables achieved comparable results (MAE=0.81, RMSE=1.06, R²=0.50).
- Variable selection was more efficient and robust than PCA; PCA-based models performed worse.
Conclusions:
- Machine learning models can moderately and practically predict SCA scores.
- Variable selection enhances model efficiency and robustness over PCA.
- Predictive tools can support earlier, more efficient quality monitoring in specialty coffee production.
