Filter Feature Selection for Detecting Mixture, Total Phenol, and pH of Civet Coffee

Civet coffee, a highly valued specialty coffee, is susceptible to adulteration with regular coffee, resulting in economic losses and consumer fraud. This study investigates the potential of electrical spectroscopy as a non-destructive technique for detecting civet coffee adulteration. We analyzed the bioelectrical properties of civet coffee beans and their mixtures with regular coffee, focusing on impedance parameters (Z, Lp, Ls, Rp, Rs) as potential indicators of adulteration. Two machine learning models, Artificial Neural Network (ANN) and Random Forest, were trained and evaluated using Mean Squared Error (MSE) validation to identify the most informative features for predicting mixture composition, total phenol content, and pH. The findings demonstrate that impedance parameters, particularly Z, consistently exhibited high feature importance scores across different attribute evaluators and search methods. The optimal model, an ANN with a correlation attribute evaluator and ranker search method, achieved an MSE validation of 0.0479, indicating strong predictive accuracy. These results suggest that electrical spectroscopy, coupled with machine learning, offers a promising approach for developing automated, non-invasive methods for detecting civet coffee adulteration, thereby protecting consumers and ensuring the integrity of the specialty coffee market.