Jakub Wroblewski

Rough set algorithms in classification problem

We we present some algorithms, based on rough set theory, that can be used for the problem of new cases classification. Most of the algorithms were implemented and included in Rosetta system [43]. We present several methods for computation of decision rules based on reducts. We discuss the problem of real value attribute discretization for increasing the performance of algorithms and quality of decision rules. Finally we deal with a problem of resolving conflicts between decision rules classifying a new case to different categories (classes). Keywords: knowledge discovery, rough sets, classification algorithms, reducts, decision rules, real value attribute discretization

A New Version of Rough Set Exploration System

We introduce a new version of the Rough Set Exploration System - a software tool featuring a library of methods and a graphical user interface supporting variety of rough-set-based computations. Methods, features and abilities of the implemented software are discussed and illustrated with a case study in data analysis.

Brighthouse: an analytic data warehouse for ad-hoc queries

Brighthouse is a column-oriented data warehouse with an automatically tuned, ultra small overhead metadata layer called Knowledge Grid, that is used as an alternative to classical indexes. The advantages of column-oriented data storage, as well as data compression have already been well-documented, especially in the context of analytic, decision support querying. This paper demonstrates additional benefits resulting from Knowledge Grid for compressed, column-oriented databases. In particular, we explain how it assists in query optimization and execution, by minimizing the need of data reads and data decompression.

Genetic Algorithms in Decomposition and Classification Problems

Some combinatorical problems concerned with using rough set theory in knowledge discovery (KD) and data analysis can be successfully solved using genetic algorithms (GA) — a sophisticated, adaptive search method based on the Darwinian principle of natural selection (see [4], [6]). These problems are frequently NP-hard, as in case of reducts or templates finding (see [12]), and there is no fast and reliable way to solve them in deterministic way.

Order based genetic algorithms for the search of approximate entropy reducts

We use entropy to extend the rough set based notion of a reduct. We show that the order based genetic algorithms, applied to the search of classical decision reducts, can be used in exactly the same way in case of extracting optimal approximate entropy reducts from data.

Theoretical Foundations of Order-Based Genetic Algorithms

A lot of research on genetic algorithms theory is concentrated on classical, binary case. However, there are many other types of useful genetic algorithms (GA), e.g. tree-based (genetic programming), or order-based ones. This paper shows, that many of classical results can be transferred into the order-based GAs. The analysis includes the Schema Theorem and Markov chain modelling of order-based GA.

Covering with Reducts - A Fast Algorithm for Rule Generation

In a rough set approach to knowledge discovery problems, a set of rules is generated basing on training data using a notion of reduct. Because a problem of finding short reducts is NP-hard, we have to use several approximation techniques. A covering approach to the problem of generating rules based on information system is presented in this article. A new, efficient algorithm for finding local reducts for each object in data table is described, as well as its parallelization and some optimization notes. A problem of working with tolerances in our algorithm is discussed. Some experimental results generated on large data tables (concerned with real applications) are presented.

Application of Temporal Descriptors to Musical Instrument Sound Recognition

An automatic content extraction from multimedia files is recently being extensively explored. However, an automatic content description of musical sounds has not been broadly investigated and still needs an intensive research. In this paper, we investigate how to optimize sound representation in terms of musical instrument recognition purposes. We propose to trace trends in the evolution of values of MPEG-7 descriptors in time, as well as their combinations. Described process is a typical example of KDD application, consisting of data preparation, feature extraction and decision model construction. Discussion of efficiency of applied classifiers illustrates capabilities of possible progress in the optimization of sound representation. We believe that further research in this area would provide background for an automatic multimedia content description.

Rough Set Approach to the Survival Analysis

Application of rough set based tools to the post-surgery survival analysis is discussed. Decision problem is defined over data related to the head and neck cancer cases, for two types of medical surgeries. The task is to express the differences between expected results of these surgeries and to search for rules discerning different survival tendencies. The rough set framework is combined with the Kaplan-Meier product estimation and the Coxs proportional hazard modeling.

Roughfication of numeric decision tables: the case study of gene expression data

We extend the standard rough set-based approach to be able to deal with huge amounts of numeric attributes versus small amount of available objects. We transform the training data using a novel way of non-parametric discretization, called roughfication (in contrast to fuzzification known from fuzzy logic). Given roughfied data, we apply standard rough set attribute reduction and then classify the testing data by voting among the obtained decision rules. Roughfication enables to search for reducts and rules in the tables with the original number of attributes and far larger number of objects. It does not require expert knowledge or any kind of parameter tuning or learning. We illustrate it by the analysis of the gene expression data, where the number of genes (attributes) is enormously large with respect to the number of experiments (objects).