About Some Specific Limitations of Data Mining Application

Task. In modern market conditions, economic entities solve issues related to forecasting, making timely management decisions, on which the effective conduct of business depends. To make managerial decisions, information is used that comes from various sources (external and internal) and not only in large volume, but also in different types. In order for the decision to be optimal, the information used must be accurate, high-quality, purified and transformed from the influence of various factors. To obtain such information, modern methods of «excavation» of data (Data Mining) are used, which make it possible to reveal patterns and relationships hidden in them. In this regard, the task of studying the fundamental limitations inherent in the methods of data excavation, which has not yet been sufficiently studied, is topical. The article discusses typical data mining tasks and identifies common limitations used in solving data analysis methods. Examples of restrictions on the generality of regularities revealed on the basis of data analysis are given. Model. The article explores various methods of «digging» data for solving typical problems of data classification and clustering, identifying associations and sequences, which allows for forecasting, analysis of deviations and visualization of data in the sections required by the user for making management decisions. Conclusions. The results obtained by the authors suggest that Data Mining methods should be used with great caution regarding the prospects and breadth of their capabilities. In particular, the authors' research revealed that when using them, it is necessary to take into account the level of aggregation of meaningfully heterogeneous data into indicators that form the information base of analytical models. Practical importance. The practical importance of the study lies in the fact that it shows the possibility of obtaining ambiguous results when using different methods for solving the same problem, which in turn leads to problems associated with the objectification of the results obtained. To this end, it is necessary to develop formal-logical tools for processing Big Data, strengthening the correspondence of formal models to their biological prototype.

Big data, information technology, data mining, classification, clustering, mathematical statistics, neural networks