A machine learning approach to sustainable land use: Integrating Markov chains and XGBoost for black galingale cultivation in Thailand

This study examined the historical and economic potential of black galingale (Kaempferia parviflora) cultivation in Thailand. Using a novel hybrid machine learning approach—Markov chain modeling for temporal dynamics and XGBoost for predictive classification—we generated unbiased suitability labels and achieved highly accurate classification. The model identified 22 districts across 10 provinces as highly or moderately suitable for cultivation. Despite challenges in soil quality such as poor root penetration, unfavorable soil texture and high jarosite levels, these areas were identified as viable using spatial data and probabilistic modeling. Feature importance analysis revealed adoption percentage, jarosite depth, and elevation as key drivers of suitability. Findings aligned with Thailand's national herbal development agenda and provided a data-driven foundation for sustainable land-use policies.