[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Subscribe]
DM: Datamining AlgorithmFrom: H. Mark Hubey Date: Fri, 02 Jun 2000 18:37:18 -0400
I just put a .pdf file on datamining on my website: http://www.csam.montclair.edu/~hubey/Curse.pdf The title has words relating to "Curse of Dimensionality". I submitted the paper to the Datamining and Knowledge Discovery Journal. Here is the abstract: A complete and unified method combining Boolean Algebra, fuzzy logic, modified Karnaugh maps, neural network type training and nonlinear transformation to create a mathematical system which can be thought of as a multiplicative (or logical-AND) neural network that can be customized to recognize various types of data clustering. The method can thus be used for (1) displaying high dimensional data, especially very large datasets, (2) recognizing patterns and clusters, with the level of approximation controllable by the dataminer (3) approximating the patterns in data to various degrees, (4) preliminary analysis for determining the number of outputs of the novel neural network shown in this manuscript, (5) creating an unsupervised learning network (of the multiplicative or AND kind) that can be used to specialize itself to clustering large amounts of high-dimensional data, and finally, (6) reducing high dimensional data to basically three-dimensions for intuitive comprehension by wrapping the data on a torus. The method can easily be extended to include vector time series. The natural space for high dimensional data using the natural Hamming metric is a torus. The specifically constructed novel neural network can then be trained or fine-tuned using machine-learning algorithms on the original data or the approximated/normalized data. Furthermore we can determine the dimensionality of the phenomena that the data represent. Enjoy. Comments welcome. -- Regards, Mark hubeyh@mail.montclair.edu .-.-.-.-.-.-. http://www.csam.montclair.edu/~hubey
|
MHonArc
2.2.0