Predicting Compatibility of Cultivars in Grafting Processes Using Kernel Methods and Collaborative Filtering.

Viticulture is the cultivation and harvesting of grapes for use in the production of juices, wines and other derivatives, with great socioeconomic importance. Grafting techniques have been applied to increase productivity and quality in the sector, but the process of finding compatible cultivars is slow and costly. Although Machine Learning (ML) methods have already been applied in several applications in agriculture, their use to support grafting processes is still very scarce. This work investigates ML-based recommender systems to address the problem of scion and rootstock compatibility in grafting processes in viticulture. In the experiments, collaborative filtering algorithms and kernel-based methods were evaluated on a dataset of 251 rated interactions, reaching a F1-score of approximately 96% for the best model. The results indicated advantages of kernel-based models over standard collaborative filtering models, as well as demonstrated the feasibility of a decision support tool to guide the choice of the best cultivars for grafting.