Final Remarks and Challenging Problems

Abstract

In this book, we studied the problem of data stream mining. Recently, it became a very important and challenging issue of computer science research. The reason is the enormous growth of data amounts generated in various areas of human activities. Data streams [1, 2, 3] are potentially of infinite size and often arrive at the system with very high rates. Therefore, it is not possible to store all the data in memory. Appropriate algorithms should use some synopsis structures to compress the information gathered from the past data. Moreover, data stream mining algorithms should be fast enough. Most often they have an incremental nature, i.e. each data element is processed at most once. Alternatively, the data stream can be analyzed in a block-based manner. Another feature of data streams is that the underlying data distribution may change over time. It is known in the literature as ‘concept drift’ [4, 5]. A good data stream mining method should be able to react to different types of changes. In this book, we studied various data stream mining algorithms. We focused on three groups of methods, based on decision trees, probabilistic neural networks, and ensemble methods. A separate part of the book was devoted to each group.