Teresa Mroczek

A comparison of six approaches to discretization: a rough set perspective

We present results of extensive experiments performed on nine data sets with numerical attributes using six promising discretization methods. For every method and every data set 30 experiments of ten-fold cross validation were conducted and then means and sample standard deviations were computed. Our results show that for a specific data set it is essential to choose an appropriate discretization method since performance of discretization methods differ significantly. However, in general, among all of these discretization methods there is no statistically significant worst or best method. Thus, in practice, for a given data set the best discretization method should be selected individually.

Rules from Belief Networks: A Rough Set Approach

A new version of the Belief SEEKER software that incorporates some aspects of rough set theory is discussed in this paper. The new version is capable of generating certain belief networks (for consistent data) and possible belief networks (for inconsistent data). Then, both types of networks can be readily converted onto respective sets of production rules, which includes both certain and/or possible rules. The new version or broadly speaking-methodology, was tested in mining the melanoma database for the best descriptive attributes of skin illness. It was found, that both types of knowledge representation, can be readily used for classification of melanocytic skin lesions.

A Comparison of Four Approaches to Discretization Based on Entropy

We compare four discretization methods, all based on entropy: the original C4.5 approach to discretization, two globalized methods, known as equal interval width and equal frequency per interval, and a relatively new method for discretization called multiple scanning using the C4.5 decision tree generation system. The main objective of our research is to compare the quality of these four methods using two criteria: an error rate evaluated by ten-fold cross-validation and the size of the decision tree generated by C4.5. Our results show that multiple scanning is the best discretization method in terms of the error rate and that decision trees generated from datasets discretized by multiple scanning are simpler than decision trees generated directly by C4.5 or generated from datasets discretized by both globalized discretization methods.

Classification and synthesis of medical images in the domain of melanocytic skin lesions

In this paper, computer-aided diagnosing and classification of melanocytic skin lesions is dealt with and briefly described. The main goal of our research was to elaborate an entirely new version of the previously developed decision support system, available now in the Internet. Its functionality and structure is here in short reported. In the current version of the system, five learning models are implemented to supply five independent results. Then, a special voting algorithm is applied to select the correct class (concept) of the diagnosed skin lesion. Additionally, some algorithms to synthesize static images of melanocytic skin lesions are briefly outlined.

Data mining analysis of granular bed caking during hop extraction

Our objective was to investigate an influence of some factors contributing to the bed caking tendency during the hop extraction process. It is important to keep the process free of bed caking since it prolongs extraction. In our research, three data sets describing the extraction process were used for knowledge discovery using rule induction and generation of belief networks. Experts analyzed discovered knowledge from the view point of potential applications. Finally, an error rate for all used methods was estimated using ten-fold cross validation.

Characteristic Sets and Generalized Maximal Consistent Blocks in Mining Incomplete Data

Mining incomplete data using approximations based on characteristic sets is a well-established technique. It is applicable to incomplete data sets with a few interpretations of missing attribute values, e.g., lost values and “do not care” conditions. Typically, probabilistic approximations are used in the process. On the other hand, maximal consistent blocks were introduced for incomplete data sets with only “do not care” conditions, using only lower and upper approximations. In this paper we introduce an extension of the maximal consistent blocks to incomplete data sets with any interpretation of missing attribute values and with probabilistic approximations. Additionally, we present results of experiments on mining incomplete data using both characteristic sets and maximal consistent blocks, using lost values and “do not care” conditions. We show that there is a small difference in quality of rule sets induced either way. However, characteristic sets can be computed in polynomial time while computing maximal consistent blocks is associated with exponential time complexity.

Belief Rules vs. Decision Rules: A Preliminary Appraisal of the Problem

An in-house developed computer program system Belief SEEKER, capable to generate belief networks and to convert them into respective sets of belief rules, was applied in mining the melanoma database. The obtained belief rules were compared with production rules generated by LERS system. It was found, that belief rules can be presumably treated as a generalization of standard IF...THEN rules.

Classification of Medical Images in the Domain of Melanoid Skin Lesions

In this paper we discuss computer-aided diagnosing and classification of melanoid skin lesions. The main goal of our research was to elaborate and to promote via Internet a new skin lesion diagnostic computer system. Its functionality and structure is described briefly in this report. In the current version of the system, five learning models are implemented to simultaneously supply five independent, partial results. Then, a special evaluation and voting algorithm is applied to select the correct class (concept) of the diagnosed skin lesion.

A Comparison of Two Approaches to Discretization: Multiple Scanning and C4.5

In a Multiple Scanning discretization technique the entire attribute set is scanned many times. During every scan, the best cutpoint is selected for all attributes. The main objective of this paper is to compare the quality of two setups: the Multiple Scanning discretization technique combined with the C4.5 classification system and the internal discretization technique of C4.5. Our results show that the Multiple Scanning discretization technique is significantly better than the internal discretization used in C4.5 in terms of an error rate computed by ten-fold cross validation (two-tailed test, 5 % level of significance). Additionally, the Multiple Scanning discretization technique is significantly better than a variant of discretization based on conditional entropy introduced by Fayyad and Irani called Dominant Attribute. At the same time, decision trees generated from data discretized by Multiple Scanning are significantly simpler from decision trees generated directly by C4.5 from the same data sets.

A new machine learning tool for mining brain stroke data

This paper contains a brief description of a new computer programming tool for supervised machine learning, designed to generate production rules from data. The research tool described — named NGTS — was used for prediction of the Glasgow Outcome Scale and Rankin Scale for patients affected by severe brain damage.