A Review of Characterization Approaches for Smallholder Farmers: Towards Predictive Farm Typologies

Area of Science:

• Agricultural economics
• Machine learning applications
• Farm management systems

Background:

• Smallholder farmer characterization is crucial for targeted interventions.
• Existing methods include machine learning, participatory, and expert-based approaches, resulting in farm typologies.
• Current typologies often lack predictive power for future trends.

Purpose of the Study:

• To review and compare methods for smallholder farmer characterization.
• To identify the strengths and weaknesses of different farm typology approaches.
• To propose a hybrid methodology for developing predictive farm typologies.

Main Methods:

• Systematic literature search on ScienceDirect and Google Scholar (2007-2018).
• Analysis of 20 research articles focusing on smallholder farmer characterization methods.
• Evaluation of cluster-based (unsupervised) and supervised learning algorithms.

Main Results:

• Cluster-based algorithms are predominantly used for smallholder farmer characterization.
• Unsupervised methods show limitations in predictability and consistency.
• Supervised models are recommended for validating unsupervised models.

Conclusions:

• A hybrid approach combining unsupervised and supervised learning is recommended for robust farm typologies.
• A three-stage characterization process is proposed: comparative unsupervised analysis, robustness assessment, and predictive power evaluation.
• The proposed method was tested on smallholder dairy farmer datasets to achieve predictive farm typologies.