The specifics and the impact of defining the initial set of centroids on information analysis

Abstract

The purpose of this paper is to highlight the importance and influence of the way of defining the initial data for the realization of algorithms for classifying information. At first glance, the problem of choosing the initial set of data processing centers for information analysis purposes could be considered as simple and obvious. In fact, things are not so clear in the initial stage of data grouping, given that the choice of clustering centers, in most cases, is strongly influenced by the specificity of each formulated problem. Even if in the absolute majority of the fieldworkers this is highlighted, but there is no suggestion of a possible way, if such a one exists, to bypass this phenomenon. The very simple clustering algorithm provides us with a fairly fast convergence, which is why it is more important that the initial parameters of the process of involving this algorithm solve our problem at a respectable efficiency level, because an optimal solution, in terms of the mathematical strictures regarding the notion "optimum", in real life, does not exist. Then, for example, when we use the word like "good", it is necessary to specify in concrete, what we mean. In conclusion: for each concrete problem it is necessary to initiate a deeper study of the impact level of the phenomenon of dependence on the final result of the clustering, and, thus defining the initial data set of the algorithm, to lead us to a solution, being on as close as possible, which we could consider the most acceptable. JEL: C63, I21, I23, I25, I29.