The Language of Soil: Soil Analysis with a Machine Learning Approach

In Türkiye, rapid population growth combined with unsustainable agricultural practices threatens the sustainable use of fertile soils and poses serious risks for the agricultural sector. To address this challenge, the present study analyzes soil data from the Odunpazarı district of Eskişehir province and proposes a machine learning–based approach. First, Principal Component Analysis (PCA) was applied to reduce data dimensionality, after which the K-Means algorithm classified the soils into three clusters. These clusters revealed significant differences in physical structure, moisture, salinity, and mineral composition, thereby providing a robust basis for further modeling. Building on this foundation, supervised machine learning models were developed and their performances compared. Logistic Regression achieved the highest accuracy (98.9%), followed by Decision Tree (97.8%), Random Forest (97.2%), and K-Nearest Neighbors (91.7%). The findings demonstrate that machine learning algorithms can reliably predict soil group membership and generate valuable insights for regional soil productivity analysis. Overall, the study highlights the effectiveness of data-driven methods in supporting sustainable agricultural planning and offers an integrative model that can guide future applications in precision agriculture.