In recent years, various methods have been investigated to improve classification accuracy. One prominent approach is ensemble learning, which involves constructing a model by integrating multiple models. Ensemble learning aims to consider multiple classifiers and combine them to obtain a classifier that outperforms each of them. Ensemble learning offers several advantages over a single model; it reduces the spread or dispersion of predictions and improves model performance. In this study, the parallel method within the ensemble learning framework is used. The parallel method involves performing the classification algorithms independently on the same training data and then combining the results using a decision rule. Specifically, our parallel ensemble method comprises two main steps: (i) implementing Multivariate Adaptive Regression Splines (MARS) and Random Forest (RF) algorithms separately, (ii) generating a MARS-RF model by integrating the results of the MARS and RF algorithms using a decision rule in MATLAB software. The decision rule employed in this study involves selecting the result of the algorithm with the highest producer’s accuracy for each class. The study area selected for investigation is the Bafra district of Samsun and Landsat-9 OLI-2 satellite image of September 30, 2022 was used. Considering the CORINE classification categories, six classes were created, including water, forest, urban fabric, heterogeneous agricultural areas, shrub and/or herbaceous vegetation associations, and open spaces with little or no vegetation. The performances of three classification techniques (RF, MARS and MARS-RF) were evaluated using kappa statistics. The kappa values for the RF and MARS algorithms were 0.801 and 0.818, respectively. The kappa value of the MARS-RF model, obtained by combining the results of RF and MARS based on the decision rule, was calculated as 0.823. This study shows that the ensemble learning method can improve classification accuracy. ORCID NO: 0000-0003-2625-3690
Anahtar Kelimeler: Ensemble Learning, Image Classification, Classification Accuracy, Multivariate Adaptive Regression Splines, Random Forest
|
