Application of Data Mining Methods in Agriculture – the Possible Use of Machine Learning

Farkas, Gábor – Magyar, Péter – Molnár, András – Zubor-Nemes, Anna

Keywords: data mining, matrix factorization, machine learning, agricultural digitalization, Farm Accountancy Data Network (FADN), C55

The increasing application of digital technologies in agriculture raises several closely related issues. Do we have more information just because we have more data? Do we need new methods and approaches beyond and/or alongside classic quantitative ones? In this article we try to provide a case study about what the possibilities of matrix-factorization (MF) are in case of FADN data. Our results show, that a well-structured MF model can efficiently learn using high quality dataset. One important finding is, that missing data have systematic nature in a sense, that they happen in case of almost all agents. We found that in the data used in agriculture: (1) missing data can be estimated with great accuracy; (2) one can implement data quality checks. Based on the results, it is recommended to study other data environments as well. Finally, it is important to stress, that the usability of the method strongly depends on the proper understanding of complex contiguity and the appropriate definition of the MF problem.

Full article