Improvement of classification performance evaluation criteria using hybrid clustering methods

Today, with the development of technology, the amount of data produced is increasing rapidly. In order to obtain meaningful information from the produced data, the data must be processed. Making complex data meaningful shortens the process of obtaining information and making decisions. For this reason, researchers are looking for ways to obtain meaningful information. Data mining methods, which include techniques such as classification and clustering, also include processing the data and transforming it into information by making the necessary operations to give meaning to the data. The aim of the study carried out within the scope of the thesis is to compare the performances of classification algorithms directly applied to the datasets frequently used in the literature and the performances of the proposed methods and classification algorithms, and also to determine the method with the best performance. The medium and large data sets determined in the study were first preprocessed, and then only the preprocessed data were compared with the performance evaluation criteria obtained by applying the methods in three stages: the K-means clustering method, the proposed hybrid clustering method and the classification processes. At the end of each stage, improvement rates were increased by parameter optimization. It has been observed that the hybrid method proposed in the thesis provides improvement in a significant part of the data sets.