Exploring Sensory Features To Predict Beer Ratings: A Comparative Machine Learning Study

This paper presents a machine learning approach for predicting beer ratings, applicable to other domains. XGBoost, Random Forest, SVM, and Neural Networks were evaluated using nested cross-validation. XGBoost showed the best and most stable performance. Feature engineering addressed sparsity, extreme values, and multicollinearity in ABV, Style, and mean IBU. Key predictors were Spices, Body, Fruits, mean IBU, and ABV. Our results are available online, promoting scientific transparency and bridging academia and industry.