TITLE:
Detection of Soil Chemical Profiles Using Supervised Learning: A Comparison of Ensemble Algorithms
AUTHORS:
Ahmed Babacar Sarr, Mapathé Ndiaye, Sabou Sarr, Abdoulaye Cisse, Ndiouga Camara
KEYWORDS:
Soil Classification, Geochemistry, Cluster, Geochemistry Factors, Supervised Learning Techniques
JOURNAL NAME:
Journal of Geoscience and Environment Protection,
Vol.14 No.5,
May
28,
2026
ABSTRACT: Geophysical prospecting comprises a set of methods used to measure variations in a physical field or in the Earth’s chemical potential. It plays a crucial role in subsurface exploration to characterize these heterogeneities. This enables the location of buried structures, lithologies or geological features to be determined. The aim of this study is to evaluate the effectiveness of machine learning algorithms in soil classification based on chemical properties. A series of samples was collected from various regions of Senegal and examined in the laboratory to determine their chemical compositions. Data pre-processing, ranging from cleaning missing values and handling inconsistencies to normalization and checking statistical consistency, was carried out on the data prior to the application of machine learning algorithms to generate classification reports. The results obtained following data processing show that ensemble methods, particularly random forests, are effective classifiers, with the X-gradient classifier and the bagging classifier achieving the highest classification accuracy of over 98%. These results demonstrate the relevance of using machine learning algorithms as tools for soil classification, complementing geotechnical studies. However, it should be noted that this study was conducted on samples whose chemical composition is known, and which were collected in an environment conducive to producing a specific chemical content depending on the soil type and region.