Özet:

This study aims to analyze the sensitivities of classification algorithms. As the work area, the Urla district of İzmir has been selected and the LANDSAT 8 images of the area are processed simultaneously. In the classification, artificial nerve networks (YSA), support vector machines (DVM) and decision trees (KA) algorithms were used. For classification, the tissues consisting of cities, forests, greenhouses, grapes, orange, crop and burned areas were selected. The selection of these tissues was carried out by carrying out on-site observation (in-situ) research with the GPS device. The results of the study were obtained by testing the sensitivities of classifications using the complex matrix algorithm. According to the images obtained from different seasons, when general accuracy was taken into account, the study with the interval data provided 8% more accuracy than the study with the April data. According to the algorithms, the highest accuracy rate is provided by the YSA algorithm with 85%. The KA algorithm has been identified as an algorithm that provides the lowest accuracy rate in classification, although it is advantageous in terms of time spent on the computer. DVM algorithms are advantageous in broad-space classification studies, taking into account both the time spent on the computer and the accuracy analysis. This study provides data that may be useful about the conservation and management of regional natural resources. Analysis of the sensitivities of classification algorithms can provide a basis for similar studies and can be a reference for future research.

Anahtar Kelimeler: