Efficiency of Classification Algorithms in Monitoring Land Use through Remote Sensing in Sissili Province of Burkina Faso

Abstract

Monitoring of natural resources is a major challenge that remote sensing tools help to facilitate. The Sissili province in Burkina Faso is a territory that includes significant areas dedicated for the preservation of forest resources. The development of satellite image processing tools offers an opportunity for better monitoring of these resources. The aim of this research is to evaluate the performance of classification algorithms in order to determine which one is most suitable for assessing and monitoring land use in the Sissili province. The methodology used is based on the comparative application of three classification algorithms: Maximum Likelihood, Random Forests, and Support Vector Machine. These algorithms were tested on land use units in the Municipality of Sissili, to determine their respective performance. Performance measures were based on the production of the confusion matrix for each algorithm, the calculation of overall accuracy, and the calculation of Kappa coefficient for the three algorithms. The results show that the Random Forest algorithm is the most effective, with a Kappa coefficient of 0.92 and an overall accuracy of 95.14%. This algorithm is followed by SVM with a Kappa coefficient of 0.73 and a maximum overall accuracy of 90.37%. The least effective algorithm for classifying land use units is Maximum Likelihood, with a Kappa coefficient of 0.36 and an overall accuracy of 52.95%. These results clearly demonstrate the superior effectiveness of machine learning algorithms, specifically Random Forest, in classifying land use units in the Sissili province.