Comparison of algorithms for dimensionality reduction and their application to index generation functions

Abstract

The selection of attributes is essential in knowledge discovery–it is fundamental for data mining algorithms, especially for more efficient classification and prediction from data. The article examines and compares the effectiveness of reducing the multidimensionality of decision tables for three RSES, jMAF, Weka programs and developed proprietary software using standard benchmarks. Next, the application of the reduction algorithm on the index generation function implementation is shown. Index generation functions are useful in the distribution of IP addresses, virus scanning or undesired data detection. In this paper an original method for the efficient implementation of index generation functions has been used. It is a multilevel logic synthesis scheme based on argument reduction that can be applied for novel heterogeneous programmable structures. Furthermore, taking into account the reduction algorithm, the discussed method is well suited to the ROM-based synthesis of index generation functions.