Optimization of pour-over coffee extraction variables using reinforcement learning

Pour-over coffee brewing is influenced by multiple interdependent variables—roast level, grind size, brew ratio, extraction time, water temperature, and total dissolved solids (TDS)—that collectively determine the final flavor and quality. This study explores the optimization of these variables using reinforcement learning (RL) and compares its performance with three common machine learning models: K-nearest neighbors (KNN), decision tree, and support vector machine (SVM). The RL agent was designed to balance exploration and exploitation, aiming to maximize rewards by adjusting brewing parameters. Data were gathered from both professional baristas and homebrewers to train and evaluate the models. The RL approach achieved the highest accuracy (90.00%), precision (90.76%), recall (90.00%), and F1-score (90.08%), outperforming KNN (accuracy: 88.33%, F1-score: 88.90%), and significantly surpassing decision tree and SVM classifiers, both of which exhibited high recall (100%) but only 50.00% accuracy and 66.67% F1-score. These findings highlight RL’s superior capability for optimizing complex, interdependent variables in pour-over coffee brewing, offering a systematic and dynamic approach to improving consistency and quality.