ÇOK DEĞİŞKENLİ VERİ KÜMELERİNDE İLGİNÇ ÖRÜNTÜ TESPİTİ İÇİN BİLEŞEN ANALİZİ

In recent years, great advances have been made on the concept of data, which has become the new power source of our age. Thanks to new methods and techniques at both coding and mechanical level, tremendous speeds have been achieved in the transfering, storing, and processing of data. Thanks to those digital developments, storing even the smallest information on digital platforms has become a natural part of daily life. From family photos to health history, from commercial records to academic publications, from a comment shared on Twitter to a video shared on Youtube, data in almost every field is stored instantly in different sizes. Interesting patterns and information in stored data waiting to be revealed are the main goals of data mining. In data mining studies, the size of data is one of the biggest problems encountered. Some of the problems encountered in large-scale data are the length of the processes of structuring such data and the jams that may occur during the execution of a model to be created afterward. Many dimension reduction algorithms have been developed to overcome the problems arising from large data sizes. In this study, a new dimension reduction approach has been developed on multivariate data. This approach generally consists of pattern recognition steps based on Principal Component Analysis (PCA). The created models were applied on disjoint and balanced sub-datasets and all produced significant results at the 0.05 confidence level. Explanatory performances of the models; They are in the range of [0.819, 0.888] on the multiple R-Square scale and in the range of [0.804, 0.878] on the R-Square scale.