Creating adaptive farm typologies using Naive Bayesian classification

Summary

The applicability of statistical typologies that capture farming systems diversity in innovation and development projects would increase if their adaptability would be enhanced, so that newly encountered farms can be classified and used to update the typology. In this paper we propose Naïve Bayesian (NB) classification as a method to allocate farms to types by using only a few variables, thus allowing the addition of new entries to a typology. We show for two example datasets that the performance of NB classification is already acceptable when 50% of the original survey dataset to construct the typology is used for training the NB classifier. For our datasets, the performance of Naïve Bayesian classification was improved when probabilities for observations to belong to multiple types were used, requiring a sample size of 30% of the survey dataset. Based on the results in this paper, we argue that NB classification is a powerful and promising statistical approach to increase the adaptability and usability of farm typologies.