Évaluation comparative des algorithmes d'apprentissage automatique pour la classification des types de sols à partir de caractéristiques physico-chimiques : application de Random Forest, XGBoost, SVM et KNN

This study explores the application of machine learning algorithms for soil type classification, a crucial task for agriculture and sustainable land management, particularly in areas facing environmental challenges such as soil degradation. The research was conducted in the Djilor commune, located i...

Full description

Saved in:
Bibliographic Details
Main Authors: Mamadou Ndiaye, René Boissy, Mbagnick Faye, N’kpomé Styvince Romaric Kouao
Format: Article
Language:fra
Published: Éditions en environnement VertigO 2025-04-01
Series:VertigO
Subjects:
Online Access:https://journals.openedition.org/vertigo/48619
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849302946872819712
author Mamadou Ndiaye
René Boissy
Mbagnick Faye
N’kpomé Styvince Romaric Kouao
author_facet Mamadou Ndiaye
René Boissy
Mbagnick Faye
N’kpomé Styvince Romaric Kouao
author_sort Mamadou Ndiaye
collection DOAJ
description This study explores the application of machine learning algorithms for soil type classification, a crucial task for agriculture and sustainable land management, particularly in areas facing environmental challenges such as soil degradation. The research was conducted in the Djilor commune, located in the Sine Saloum region of Senegal, an area affected by salinization and agricultural land loss. Using physico-chemical characteristics such as texture (percentages of sand, silt, and clay), pH, organic matter, cation exchange capacity (CEC), bulk density, and water retention, this study evaluates the performance of four algorithms: Random Forest, XGBoost, SVM, and KNN. A set of 1000 random samples was used for training and testing, with cross-validation and confusion matrices to assess performance. The results show that SVM achieves the best performance with an overall accuracy of 98.85%, followed by Random Forest (97.13%) and KNN (95.40%), while XGBoost shows an accuracy of 93.68%. These results highlight the ability of the models to capture the complex relationships between the physico-chemical characteristics, although adjustments are needed to improve their ability to handle minority classes. The analysis of feature importance reveals that soil texture, particularly the percentages of sand and silt, plays a key role in soil type classification. These results highlight the relevance of machine learning as a tool for sustainable management of agricultural land and natural resources. However, the study also highlights limitations, such as the need for validation with field data and exploration of more complex models to improve the robustness and generalisability of the conclusions.
format Article
id doaj-art-e6ce36509df84820bce686840ffff7b7
institution Kabale University
issn 1492-8442
language fra
publishDate 2025-04-01
publisher Éditions en environnement VertigO
record_format Article
series VertigO
spelling doaj-art-e6ce36509df84820bce686840ffff7b72025-08-20T04:02:05ZfraÉditions en environnement VertigOVertigO1492-84422025-04-0110.4000/13raqÉvaluation comparative des algorithmes d'apprentissage automatique pour la classification des types de sols à partir de caractéristiques physico-chimiques : application de Random Forest, XGBoost, SVM et KNNMamadou NdiayeRené BoissyMbagnick FayeN’kpomé Styvince Romaric KouaoThis study explores the application of machine learning algorithms for soil type classification, a crucial task for agriculture and sustainable land management, particularly in areas facing environmental challenges such as soil degradation. The research was conducted in the Djilor commune, located in the Sine Saloum region of Senegal, an area affected by salinization and agricultural land loss. Using physico-chemical characteristics such as texture (percentages of sand, silt, and clay), pH, organic matter, cation exchange capacity (CEC), bulk density, and water retention, this study evaluates the performance of four algorithms: Random Forest, XGBoost, SVM, and KNN. A set of 1000 random samples was used for training and testing, with cross-validation and confusion matrices to assess performance. The results show that SVM achieves the best performance with an overall accuracy of 98.85%, followed by Random Forest (97.13%) and KNN (95.40%), while XGBoost shows an accuracy of 93.68%. These results highlight the ability of the models to capture the complex relationships between the physico-chemical characteristics, although adjustments are needed to improve their ability to handle minority classes. The analysis of feature importance reveals that soil texture, particularly the percentages of sand and silt, plays a key role in soil type classification. These results highlight the relevance of machine learning as a tool for sustainable management of agricultural land and natural resources. However, the study also highlights limitations, such as the need for validation with field data and exploration of more complex models to improve the robustness and generalisability of the conclusions.https://journals.openedition.org/vertigo/48619Senegalmachine learningRandom ForestXGBoostsupport vector machineK-Nearest Neighbors
spellingShingle Mamadou Ndiaye
René Boissy
Mbagnick Faye
N’kpomé Styvince Romaric Kouao
Évaluation comparative des algorithmes d'apprentissage automatique pour la classification des types de sols à partir de caractéristiques physico-chimiques : application de Random Forest, XGBoost, SVM et KNN
VertigO
Senegal
machine learning
Random Forest
XGBoost
support vector machine
K-Nearest Neighbors
title Évaluation comparative des algorithmes d'apprentissage automatique pour la classification des types de sols à partir de caractéristiques physico-chimiques : application de Random Forest, XGBoost, SVM et KNN
title_full Évaluation comparative des algorithmes d'apprentissage automatique pour la classification des types de sols à partir de caractéristiques physico-chimiques : application de Random Forest, XGBoost, SVM et KNN
title_fullStr Évaluation comparative des algorithmes d'apprentissage automatique pour la classification des types de sols à partir de caractéristiques physico-chimiques : application de Random Forest, XGBoost, SVM et KNN
title_full_unstemmed Évaluation comparative des algorithmes d'apprentissage automatique pour la classification des types de sols à partir de caractéristiques physico-chimiques : application de Random Forest, XGBoost, SVM et KNN
title_short Évaluation comparative des algorithmes d'apprentissage automatique pour la classification des types de sols à partir de caractéristiques physico-chimiques : application de Random Forest, XGBoost, SVM et KNN
title_sort evaluation comparative des algorithmes d apprentissage automatique pour la classification des types de sols a partir de caracteristiques physico chimiques application de random forest xgboost svm et knn
topic Senegal
machine learning
Random Forest
XGBoost
support vector machine
K-Nearest Neighbors
url https://journals.openedition.org/vertigo/48619
work_keys_str_mv AT mamadoundiaye evaluationcomparativedesalgorithmesdapprentissageautomatiquepourlaclassificationdestypesdesolsapartirdecaracteristiquesphysicochimiquesapplicationderandomforestxgboostsvmetknn
AT reneboissy evaluationcomparativedesalgorithmesdapprentissageautomatiquepourlaclassificationdestypesdesolsapartirdecaracteristiquesphysicochimiquesapplicationderandomforestxgboostsvmetknn
AT mbagnickfaye evaluationcomparativedesalgorithmesdapprentissageautomatiquepourlaclassificationdestypesdesolsapartirdecaracteristiquesphysicochimiquesapplicationderandomforestxgboostsvmetknn
AT nkpomestyvinceromarickouao evaluationcomparativedesalgorithmesdapprentissageautomatiquepourlaclassificationdestypesdesolsapartirdecaracteristiquesphysicochimiquesapplicationderandomforestxgboostsvmetknn