Taste Matters: Machine Learning Models for Context-Aware Recipe Prediction

Abstract

Taste has always been a decisive factor in food and beverage preparation. Yet, in times of increasing ecological awareness, optimizing recipes requires balancing subjective user satisfaction with measurable sustainability goals. Coffee, one of the most widely consumed beverages, provides a particularly relevant case: small changes in the Coffee-to-Water Ratio (C2WR) not only influence taste perception but also have a measurable impact on the environmental footprint. Building on previous work that established a universal architecture for context-aware food and beverage preparation systems (CONFES) and developed a large-scale data acquisition framework for a context-aware coffee machine, this paper extends the research toward machine learning modelling approaches capable for prediction of recipe parameters like C2WR. Tree-based ensemble models, such as Random Forest, Gradient Boosting and AdaBoost explained a higher proportion of variance (R² = 61.5%) compared to Neural Networks, k-Nearest Neighbour, and Support Vector Machines.

Keywords: Context-aware systems, Coffee-to-Water Ratio (C2WR), Recipe optimization, Machine learning, Taste perception, Sustainability, Random Forest, Gradient Boosting, AdaBoost.

DOI: 10.54941/ahfe1007177