Been-Chian Chien

Learning a coverage set of maximally general fuzzy rules by rough sets

Abstract Expert systems have been widely used in domains where mathematical models cannot be easily built, human experts are not available or the cost of querying an expert is high. Machine learning or data mining can extract desirable knowledge or interesting patterns from existing databases and ease the development bottleneck in building expert systems. In the past we proposed a method [Hong, T.P., Wang, T.T., Wang, S.L. (2000). Knowledge acquisition from quantitative data using the rough-set theory. Intelligent Data Analysis (in press).], which combined the rough set theory and the fuzzy set theory to produce all possible fuzzy rules from quantitative data. In this paper, we propose a new algorithm to deal with the problem of producing a set of maximally general fuzzy rules for coverage of training examples from quantitative data. A rule is maximally general if no other rule exists that is both more general and with larger confidence than it. The proposed method first transforms each quantitative value into a fuzzy set of linguistic terms using membership functions and then calculates the fuzzy lower approximations and the fuzzy upper approximations. The maximally general fuzzy rules are then generated based on these fuzzy approximations by an iterative induction process. The rules derived can then be used to build a prototype knowledge base in a fuzzy expert system.

Mining Fuzzy Multiple-Level Association Rules from Quantitative Data

Machine-learning and data-mining techniques have been developed to turn data into useful task-oriented knowledge. Most algorithms for mining association rules identify relationships among transactions using binary values and find rules at a single-concept level. Transactions with quantitative values and items with hierarchical relationships are, however, commonly seen in real-world applications. This paper proposes a fuzzy multiple-level mining algorithm for extracting knowledge implicit in transactions stored as quantitative values. The proposed algorithm adopts a top-down progressively deepening approach to finding large itemsets. It integrates fuzzy-set concepts, data-mining technologies and multiple-level taxonomy to find fuzzy association rules from transaction data sets. Each item uses only the linguistic term with the maximum cardinality in later mining processes, thus making the number of fuzzy regions to be processed the same as the number of original items. The algorithm therefore focuses on the most important linguistic terms for reduced time complexity.

Classifier design with feature selection and feature extraction using layered genetic programming

This paper proposes a novel method called FLGP to construct a classifier device of capability in feature selection and feature extraction. FLGP is developed with layered genetic programming that is a kind of the multiple-population genetic programming. Populations advance to an optimal discriminant function to divide data into two classes. Two methods of feature selection are proposed. New features extracted by certain layer are used to be the training set of next layers populations. Experiments on several well-known datasets are made to demonstrate performance of FLGP.

Learning discriminant functions with fuzzy attributes for classification using genetic programming

Abstract Classification is one of the important tasks in developing expert systems. Most of the previous approaches for classification problem are based on classification rules generated by decision trees. In this paper, we propose a new learning approach based on genetic programming to generate discriminant functions for classifying data. An adaptable incremental learning strategy and a distance-based fitness function are developed to improve the efficiency of genetic programming-based learning process. We first transform attributes of objects into fuzzy attributes and then a set of discriminant functions is generated based on the proposed learning procedure. The set of derived functions with fuzzy attributes gives high accuracy of classification and presents a linear form. Hence, the functions can be transformed into inference rules easily and we can use the rules to provide the building of rule base in an expert system.

Designing a classifier by a layered multi-population genetic programming approach

This paper proposes a method called layered genetic programming (LAGEP) to construct a classifier based on multi-population genetic programming (MGP). LAGEP employs layer architecture to arrange multiple populations. A layer is composed of a number of populations. The results of populations are discriminant functions. These functions transform the training set to construct a new training set. The successive layer uses the new training set to obtain better discriminant functions. Moreover, because the functions generated by each layer will be composed to a long discriminant function, which is the result of LAGEP, every layer can evolve with short individuals. For each population, we propose an adaptive mutation rate tuning method to increase the mutation rate based on fitness values and remaining generations. Several experiments are conducted with different settings of LAGEP and several real-world medical problems. Experiment results show that LAGEP achieves comparable accuracy to single population GP in much less time.

Mining from incomplete quantitative data by fuzzy rough sets

Machine learning can extract desired knowledge from existing training examples and ease the development bottleneck in building expert systems. Most learning approaches derive rules from complete data sets. If some attribute values are unknown in a data set, it is called incomplete. Learning from incomplete data sets is usually more difficult than learning from complete data sets. In the past, the rough-set theory was widely used in dealing with data classification problems. Most conventional mining algorithms based on the rough-set theory identify relationships among data using crisp attribute values. Data with quantitative values, however, are commonly seen in real-world applications. In this paper, we thus deal with the problem of learning from incomplete quantitative data sets based on rough sets. A learning algorithm is proposed, which can simultaneously derive certain and possible fuzzy rules from incomplete quantitative data sets and estimate the missing values in the learning process. Quantitative values are first transformed into fuzzy sets of linguistic terms using membership functions. Unknown attribute values are then assumed to be any possible linguistic terms and are gradually refined according to the fuzzy incomplete lower and upper approximations derived from the given quantitative training examples. The examples and the approximations then interact on each other to derive certain and possible rules and to estimate appropriate unknown values. The rules derived can then serve as knowledge concerning the incomplete quantitative data set.

A color image segmentation approach based on fuzzy similarity measure

We propose an image segmentation scheme based on a fuzzy color similarity measure to segment out meaningful objects in an image according to human perception. The proposed method first defines a set of fuzzy colors based on the HLS color coordinate space. Each pixel in an image is represented by a set of fuzzy colors that are the most similar colors in the color palette selected by humans. Then, a fuzzy similarity measure is developed for evaluating the similarity of fuzzy colors between two pixels. We recursively merge adjacent pixels to form meaningful objects by the fuzzy similarity measure until there is no similar color between adjacent pixels. Experiments demonstrate that the proposed method can extract meaningful objects from images effectively.

An efficient clustering algorithm for mining fuzzy quantitative association rules

Mining association rules on categorical data has been discussed widely. It is a relatively difficult problem in the discovery of association rules from numerical data, since the reasonable intervals for unknown numerical attributes or quantitative data may not be discriminated easily. We propose an efficient hierarchical clustering algorithm based on variation of density to solve the problem of interval partition. We define two main characteristics of clustering numerical data: relative inter-connectivity and relative closeness. By giving a proper parameter, /spl alpha/, to determine the importance between relative closeness and relative inter-connectivity, the proposed approach can generate a reasonable interval automatically for the user. The experimental results show that the proposed clustering algorithm can have good performance on both clustering results and speed.

A two-level relevance feedback mechanism for image retrieval

Content-based image retrieval (CBIR) is a group of techniques that analyzes the visual features (such as color, shape, texture) of an example image or image subregion to find similar images in an image database. Relevance feedback is often used in a CBIR system to help users express their preference and improve query results. Traditional relevance feedback relies on positive and negative examples to reformulate the query. Furthermore, if the system employs several visual features for a query, the weight of each feature is adjusted manually by the user or system predetermined and fixed by the system. In this paper we propose a new relevance feedback model suitable for medical image retrieval. The proposed method enables the user to rank the results in relevance order. According to the ranking, the system can automatically determine the importance ranking of features, and use this ranking to automatically adjust the weight of each feature. The experimental results show that the new relevance feedback mechanism outperforms previous relevance feedback models.

Learning coverage rules from incomplete data based on rough sets

In this paper, we deal with the problem of producing a set of certain and possible rules for coverage of incomplete data sets based on rough sets. All the coverage rules gathered together can cover all the given training examples. Unknown values are first assumed to be any possible values and are gradually refined according to the incomplete lower and upper approximations derived from the given incomplete training examples. One of the best equivalence classes in incomplete lower or upper approximations is chosen according to some criteria. The objects covered by the incomplete equivalence class are then removed from the incomplete training set. The same procedure is repeated to find the coverage set of rules. The training examples and the approximations then interact on each other to find the maximally general coverage rules and to estimate appropriate unknown values. The rules derived can then be used to build a prototype knowledge base.