Developing algorithm for determining initial centroid points with orbit iteration in K-means clustering method of data mining

The most basic idea of data mining is to reach the desired information about the data set by revealing the sub-information of each object in the data set. This sub-info is revealed by various mathematical and statistical methods in data mining One of the study areas of data mining is clustering and it is known as unsupervised learning in the published literature. The main purpose of clustering techniques is to divide datasets into any number of clusters or groups. K-Means is the most commonly used method in clustering techniques. As with any method, this method has its shortcomings too. The most well-known shortcoming of the K-Means method is that it works with the selection of random initial centroids. It is a known fact that this situation negatively affects clustering success. Because it produces different results each time it is run. Moreover it makes the method not a deterministic one. In this Ph.D. dissertation, it is aimed to remove the non-deterministic nature of the K-Means method and to increase the clustering success. For this purpose, firstly, the literature on clustering methods were analyzed. Then, studies to find the initial center of gravity used in clustering were figured out. With the help of this information, a new initial centroid method based on orbit iterations of data points is designed. In addition, this new method has been coded in the Python programming language and has been tested on artificial and real datasets. The results obtained from the method were compared with the results of other studies in the literature. According to the findings, the designed method was found to be successful.