Abstract

This article conducts a systematic literature review of bee honey quality and yield estimation using machine learning in the Web of Science database. The review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses methodology, creating a workflow that includes the planning stage, execution of the review, and reporting of the results. Its description ensures this study's replicability. The Web of Science search retrieved 352 articles. After screening the titles and abstracts and reviewing the full texts, the set is reduced to 105 studies, of which 94 address honey quality and only 11 focus on yield. In general, China is the country with the most honey quality studies, and publications in this area appear mainly in food science and spectrochemical journals. Concerning honey yield, the main contributors are countries such as Australia and Turkey; these studies are published chiefly in agricultural and computer science journals. The machine learning algorithms most frequently applied to honey quality are Principal Component Analysis, Support Vector Machine, and shallow neural networks, used primarily with spectroscopic data and microscopic pollen images. The recent use of mass spectrometry in this field is also noted. For honey yield, the most common machine learning methods belong to the ensemble and boosting families, combined with data from hive sensors, climatic records, and hive yield or weight measurements. This review also highlights the use and potential of deep learning methods, explainable artificial intelligence techniques, and time series models. Finally, it identifies readily accessible data sources for future research on honey quality and yield: microscopic pollen image and hyperspectral honey image databases for quality studies, and in-hive sensor and climate data for yield investigations.