Application of Multiple Unsupervised Models to Validate Clusters Robustness in Characterizing Smallholder Dairy...

Area of Science:

• Agricultural Economics
• Data Science
• Animal Science

Background:

• Smallholder dairy systems exhibit significant heterogeneity, hindering effective service delivery and technology adoption.
• Homogenous farmer groups are essential for targeted interventions to boost productivity and profitability.

Purpose of the Study:

• To evaluate the robustness of different unsupervised learning algorithms for defining smallholder dairy production clusters.
• To identify the most effective clustering approach for diverse farming systems in Ethiopia and Tanzania.

Main Methods:

• Utilized data from 8179 smallholder dairy farms in Ethiopia and Tanzania.
• Selected 35 key variables using Principal Component Analysis and expert knowledge.
• Compared K-means, fuzzy clustering, and Self-Organizing Maps (SOM) for grouping consistency and predictive accuracy.

Main Results:

• Fuzzy clustering yielded the highest predictive power for Ethiopian data (77% milk yield, 48% milk sales).
• Self-Organizing Maps performed best for Tanzanian data.
• Cluster membership reallocation varied significantly, with fuzzy (34%) and K-means (15%) showing higher/lower instability in Ethiopia.

Conclusions:

• The optimal unsupervised learning algorithm for defining dairy production clusters is country-specific.
• Generalizing clustering models across different countries and production systems is challenging due to unique data characteristics.
• Robust cluster definition is crucial for tailored interventions in smallholder agriculture.