Search

Integrating the RBAC model in object-oriented systems is a natural way to describe authorization policies. We extend the RBAC model for access control in object-oriented systems in the context of the Access Control List. In this paper, we discuss access control issues categorizing three cases: subject to object, inter-objects, and intra object. It may be desirable in some applications to have a fine-grained access control at the level of the individual attributes or the methods of an object. We also demonstrate how access control decisions are made using $\mathcal{ALCQ}$ language, a family member of description logics. [ABSTRACT FROM AUTHOR]

The paper proposes an approach based on Rough Mereology to approximating hierarchical relationships between imprecise concepts in knowledge representation systems. The approach employs Interval Analysis to capture the imprecision caused by the granularity of knowledge. Interval rough inclusion functions are defined. It is demonstrated that they can be effectively used to compute the IS-A relationships by measuring the inclusion of one approximated concept into another. It is shown that the functions are superior to the previously suggested in literature. [ABSTRACT FROM AUTHOR]

Rough neural networks aim at hierarchical construction of compound concepts. Although the structure of such concepts is assumed to be more complicated than numbers in case of standard feedforward neural networks, some mechanisms can be generalized to achieve efficient propagation and learning. One of possible generalizations, called the normalizing neural networks, enables to propagate vectors instead of single signals. Neurons take form of multidimensional functions, which model cross-dependencies among importance of particular vector components. In this way, we are able to represent some types of compound concepts using relatively simple neural network structure. As an illustration, we consider the case study related to the task of magnetic resonance images' segmentation. We put a special emphasis on how the nature of objects and attributes in a given decision system influences the network's architecture. We also compare our model to other rough-neural approaches. [ABSTRACT FROM AUTHOR]

At first, we discuss the basic structure of the fuzzy system as a simple yet powerful fuzzy modeling technique. Neural networks and fuzzy logic models are based on very similar underlying mathematics. The similarity between RBF networks and fuzzy models is noted in detail. Then, we propose the extension of RBF neural networks by the cloud model. Time series approximation and prediction by applying RBF neural networks or fuzzy models and comparisons between the various types of RBF networks and statistical models are discussed at length. [ABSTRACT FROM AUTHOR]

We propose a health recommendation system architecture using rough sets, survival analysis approaches and rule-based expert systems. Our main goal is to recommend clinical examinations for patients or physicians from patients' self reported data. Such data will be treated as condition attributes, while survival time from a follow-up study will be treated as the target function. We have amalgamated rough set theory, relational databases, statistics, soft computing and several pertinent techniques to generate a hybrid intelligent system for survival analysis. This study represents the completion of our system by adding a recommendation module. [ABSTRACT FROM AUTHOR]

This paper presents a method of target tracking for a robotic vision system employing reinforcement learning with feedback based on average rough coverage performance values. The application is for a line-crawling inspection robot (ALiCE II, the second revision of Automated Line Crawling Equipment) designed to automate the inspection of hydro electric transmission lines and related equipment. The problem considered in this paper is how to train the vision system to track targets of interest and acquire useful images for further analysis. To train the system, two versions of Watkins' Q-learning were implemented, the classical single-step version and a modified strain using an approximation space-based form of what we term rough feedback. The robot is briefly described along with experimental results for the two forms of the Q-learning control algorithm. The contribution of this article is an introduction to a modified version of Q-learning control with rough feedback to monitor and adjust the learning rate during target tracking. [ABSTRACT FROM AUTHOR]

Stock market time series are inherently noisy. Although support vector machine has the noise-tolerant property, the noised data still affect the accuracy of classification. Compared with other studies only classify the movements of stock market into up-trend and down-trend which does not concern the noised data, this study uses wavelet soft-threshold de-noising model to classify the noised data into stochastic trend. In the experiment, we remove the stochastic trend data from the SSE Composite Index and get de-noised training data for SVM. Then we use the de-noised data to train SVM and to forecast the testing data. The hit ratio is 60.12%. Comparing with 54.25% hit ratio that is forecasted by noisy training data SVM, we enhance the forecasting performance. [ABSTRACT FROM AUTHOR]

The problem considered in this paper is how to recognize similar objects based on the detection of patterns in pairs of images. This article introduces a new form of classifier based on approximation spaces in the context of near sets for use in pattern recognition. By way of introducing the basic approach, nonlinear diffusion is used for edge detection and object contour extraction. This form of image transformation makes it possible to compare the contours of objects in pairs of images. Once the contour of an image has been identified, it is then possible to construct approximation spaces based on vectors of probe function measurements associated with selected image features. In this article, the only feature considered is contour, which leads to many contour probe functions. The contribution of this article is a new form of classifier, based on approximation spaces, for use in image pattern recognition. [ABSTRACT FROM AUTHOR]

In many real-world applications, the costs of different errors are often unequal. Therefore, the inclusion of costs into learning, also named cost-sensitive learning, has been regarded as one of the most relevant topics of future machine learning research. Rough set theory is a powerful mathematic tool dealing with inconsistent information for attribute dependence analysis, knowledge reduction and decision rule extraction. However, it is insensitive to the costs of misclassification due to the absence of a mechanism of considering the subjective knowledge. This paper discusses problems connected with introducing the subjective knowledge into rough set learning and proposes a weighted rough set approach for cost-sensitive learning. In this method, weights are employed to represent the subjective knowledge of costs and a weighted information system is defined firstly. With the introduction of weights, weighted attribute dependence analysis is carried out and an index of weighted approximate quality is given. Furthermore, weighted attribute reduction algorithm and weighted rule extraction algorithm are designed to find the reducts and rules with the consideration of weights. Based on the proposed weighted rough set, a series of comparing experimentations with several familiar general techniques on cost-sensitive learning are constructed. The results show that the approach of weighted rough set produces averagely the minimum misclassification costs and the lowest high cost errors. [ABSTRACT FROM AUTHOR]

This paper demonstrates how to employ rough set framework in order to induce JEPs in transactional data. The algorithm employs local reducts in order to generate desired JEPs and additional EPs. The number of the latter is decreased by preceding reduct computation with item aggregation. The preprocessing is reduced to graph coloring and solved with efficient classical heuristics. Our approach is contrasted with JEP-Producer, the recommended method for JEP induction. Moreover, a formal apparatus for classified transactional data has been proposed. [ABSTRACT FROM AUTHOR]

In this paper, we present a novel image digital watermarking technique based on Kernel Independent Component Analysis (KICA). Use the nice characteristic of the KICA, which can results the blind separation of nonlinearly mixed signals, the imperceptibility and robustness requirements of watermarks are fulfilled and optimized. In the proposed scheme, the watermark image is first transformed by Arnold method, and then embedded into the lowest frequency subband in DWT domain. The recovery of owner's image is turning the watermarked image into DWT domains then use KICA to extract the watermark. Finally the watermark is transformed by Arnold method again, so we can get the original watermark image. Experimental results show that the proposed method gains better performance in robustness than that of ICA with respect to traditional image processing including cropping, filtering, add noise and JPEG image compression. [ABSTRACT FROM AUTHOR]

Plasma display panel (PDP) has become popular as high-end television monitors. In PDPs, gray levels are expressed by the pulse-number modulation technique to generate gradation display. Although this method can display still image faithfully, it can yield annoying distortions when displaying moving images. In order to reduce the dynamic false contour and develop moving picture quality, we propose a rough set based effective subfield optimization technique. Simulation results show that the dynamic false contour can be effectively reduced by rough set based subfield optimization algorithm. [ABSTRACT FROM AUTHOR]

In order to address problems of information overload in digital imagery task domains we have developed an interactive approach to the capture and reuse of image context information. Our framework models different aspects of the relationship between images and domain tasks they support by monitoring the interactive manipulation and annotation of task-relevant imagery. The approach allows us to gauge a measure of a user's intentions as they complete goal-directed image tasks. As users analyze retrieved imagery their interactions are captured and an expert task context is dynamically constructed. This human expertise, proficiency, and knowledge can then be leveraged to support other users in carrying out similar domain tasks. We have applied our techniques to two multimedia retrieval applications for two different image domains, namely the geo-spatial and medical imagery domains. [ABSTRACT FROM AUTHOR]

We present a unique representation scheme for events in an area under surveillance, which provides a mechanism to analyze videos from multiple perspectives for unusual activity analysis. We propose clustering in event component spaces and define algebraic operations on these clusters to find co-occurrences of event components. A usualness measure for clusters is proposed that not only gives a measure on how usual or unusual an activity is, but also a basis for analyzing and predicting the possibly usual or unusual activities that can occur in the surveillance region. [ABSTRACT FROM AUTHOR]

A novel approach to extend the notions of definability and rough set approximations in information systems with non-equivalence relations is proposed. The upper approximation is defined as set-theoretic complement of negative region of a given concept; therefore, it does not need to be definable. Fundamental properties of new approximation operators are compared with the previous ones reported in literature. The proposed idea is illustrated within tolerance approximation spaces. In particular, granulation based on maximal preclasses is considered. [ABSTRACT FROM AUTHOR]

An ability of Pawlak's Rough Sets Theory to handle imprecision and uncertainty without any need of preliminary or additional information about analyzed data makes this theory very interesting for analyzing medical data. Using Rough Sets Theory knowledge extracted from raw data may be stored in form of decision rules. But increasing number and complexity of decision rules make their analysis and validation by domain experts difficult. In this paper we focus on this problem and propose an approach to visualize decision rules in form of decision trees. Afterwards domain experts validate transformed decision trees and compare the results with general guidelines proposed by the American College of Cardiology Foundation and the American Heart Association. [ABSTRACT FROM AUTHOR]

One of the main contributions of rough set theory to data mining is data reduction. There are three reductions: attribute (column) reduction, row reduction, and value reduction. Row reduction is merging the duplicate rows. Attribute reduction is to find important attributes. Value reduction is to reduce the decision rules to a logically equivalent minimal length. Most recent attentions have been on finding attribute reducts. Traditionally, the value reduct has been searched through the attribute reduct. This paper observes that this method may miss the best value reducts. It also revisits an old rudiment idea [11], namely, a rough set theory on high frequency data: The notion of high frequency value reduct is extracted in a bottom-up fashion without finding attribute reducts. Our method can discover concise and important decision rules in large databases, and is described and illustrated by an example. [ABSTRACT FROM AUTHOR]

This paper shows problems with combination of rule induction and attribute-oriented generalization, where if the given hierarchy includes inconsistencies, then application of hierarchical knowledge generates inconsistent rules. Then, we introduce two approaches to solve this problem, one process of which suggests that combination of rule induction and attribute-oriented generalization can be used to validate concept hiearchy. Interestingly, fuzzy linguistic variables play an important role in solving these problems. [ABSTRACT FROM AUTHOR]

In the paper, we present the relationship between loss functions and confirmation measures. We show that population minimizers for weighted loss functions correspond to confirmation measures. This result can be used in construction of machine learning methods, particularly, ensemble methods. [ABSTRACT FROM AUTHOR]

For traditional way of Web document representation in Vector Space Model, zero-valued similarity problem between vectors occurs frequently, which decreases classificatory quality when defining the relation between Web documents. In this paper, a novel Web document representation and classification approach based on rough set is proposed. Firstly, TF*IDF weighting scheme is used to assign weight values for Web document's vector. The weights of those terms which do not occur in a Web document are considered missing information. Then rough set for incomplete information is introduced to supplement loss and expand Web document representation. Through generating tolerance classes in both term space and Web document space, the missing information of Web document can be complemented by incorporating the corresponding weights of terms in tolerance classes, which extends the essential information to Web document. Finally, Web document classification algorithm is implemented. Experimental results show that the performance of the classification is greatly improved. [ABSTRACT FROM AUTHOR]

We present a generalization of a strategy, called SCIKD, proposed in [7] that allows to reduce a disclosure risk of confidential data in an information system S [10] using methods based on knowledge discovery. The method proposed in [7] protects confidential data against Rule-based Chase, the null value imputation algorithm driven by certain rules [2], [4]. This method identifies a minimal subset of additional data in S which needs to be hidden to guarantee that the confidential data are not revealed by Chase. In this paper we propose a bottom-up strategy which identifies, for each object x in S, a maximal set of values of attributes which do not have to be hidden and still the information associated with secure attribute values of x is protected. It is achieved without examining all possible combinations of values of attributes. Our method is driven by classification rules extracted from S and takes into consideration their confidence and support. [ABSTRACT FROM AUTHOR]

Clustering is a technique for grouping items in a dataset that are similar, while separating those items that are dissimilar. The use of concept lattices, from Formal Concept Analysis, for disjoint clustering is a recently studied problem. We develop an algorithm for disjoint clustering of transactional databases using concept lattices. Several heuristics are developed for tuning the support parameters used in this algorithm. Additionally, we discuss the application of this algorithm to Location Learning. In location learning, an object (for example an employee) to be tracked and localized carries an electronic tag, such as an RFID, capable of communicating with some access points that are in the range of the tag. Clustering can then be used to estimate the location of the tag given the signal strengths that can be heard. [ABSTRACT FROM AUTHOR]

This paper proposes a fuzzy model tree, so-called c-fuzzy model tree, consisting of local linear models using fuzzy cluster for data modeling. Cluster centers are calculated by fuzzy clustering method using all input and output attributes. And then, linear models are constructed at internal nodes with fuzzy membership grades between centers and input attributes. The expansion of internal node is determined by comparing the error calculated at the parent node with the sum of ones at the child nodes. On the other hand, data prediction is performed with the linear model having the highest fuzzy membership value between input attributes and cluster centers at the leaf nodes. To show the effectiveness of the proposed method, we have applied this method to real world data set. We found that the proposed method showed better performance than the widely used methods, such as model tree and artificial neural networks. [ABSTRACT FROM AUTHOR]

Density-based clustering methods are of particular interest for applications where the anticipated groups of data instances are expected to differ in size or shape, arbitrary shapes are possible and the number of clusters is not known a priori. In such applications, background knowledge about group-membership or non-membership of some instances may be available and its exploitation so interesting. Recently, such knowledge is being expressed as constraints and exploited in constraint-based clustering. In this paper, we enhance the density-based algorithm DBSCAN with constraints upon data instances - "Must-Link" and "Cannot-Link" constraints. We test the new algorithm C-DBSCAN on artificial and real datasets and show that C-DBSCAN has superior performance to DBSCAN, even when only a small number of constraints is available. [ABSTRACT FROM AUTHOR]

The hybridization of rough sets and fuzzy sets has focused on creating an end product that extends both contributing computing paradigms in a conservative way. As a result, the hybrid theory inherits their respective strengths, but also exhibits some weaknesses. In particular, although they allow for gradual membership, fuzzy rough sets are still abrupt in a sense that adding or omitting a single element may drastically alter the outcome of the approximations. In this paper, we revisit the hybridization process by introducing vague quantifiers like "some" or "most" into the definition of upper and lower approximation. The resulting vaguely quantified rough set (VQRS) model is closely related to Ziarko's variable precision rough set (VPRS) model. [ABSTRACT FROM AUTHOR]

Action rules can be seen as an answer to the question: what one can do with results of data mining and knowledge discovery? Some applications include: medical field, e-commerce, market basket analysis, customer satisfaction, and risk analysis. Action rules are logical terms describing knowledge about possible actions associated with objects, which is hidden in a decision system. Classical strategy for discovering them from a database requires prior extraction of classification rules which next are evaluated pair by pair with a goal to suggest an action, based on condition features in order to get a desired effect on a decision feature. An actionable strategy is represented as a term $r = [(\omega) \wedge (\alpha \rightarrow \beta)] \Rightarrow [\phi \rightarrow \psi]$, where ω, α, β, φ, and ψ are descriptions of objects or events. The term r states that when the fixed condition ω is satisfied and the changeable behavior (α→β) occurs in objects represented as tuples from a database so does the expectation (φ→ψ). With each object a number of actionable strategies can be associated and each one of them may lead to different expectations and the same to different re-classifications of objects. In this paper we will focus on a new strategy of constructing action rules directly from single classification rules instead of pairs of classification rules. It presents a gain on the simplicity of the method of action rules construction, as well as on its time complexity. We present A*-type heuristic strategy for discovering only interesting action rules, which satisfy user-defined constraints such as: feasibility, maximal cost, and minimal confidence. We, therefore, propose a new method for fast discovery of interesting action rules. [ABSTRACT FROM AUTHOR]

We propose a new method for evaluating the contribution that elementary conditions give to a set of rules. It is based on previous results concerning importance and interaction of elementary conditions for a confidence of a single rule. The usefulness of the method is shown on two rule discovery problems. [ABSTRACT FROM AUTHOR]

Constructing associative classifier is to use the technique of mining association rules to extract attribute-value pairs that are associated with class labels. Since too many such kinds of associations may be generated however, the existing algorithms to finding associations are usually ineffective. It is well known that rough sets theory can be used to select reducts of attributes that represent the original data set. In this paper we present an approach of combining the rough set theory, the association rules mining technique, and the covering method to construct classification rules. With a given decision table, the rough set theory is first used to find all reducts of condition attributes of the decision table. Then an adapted Apriori algorithm to mining association rules is used to find a set of associative classifications from each reduct. And third, all association classification rules are ranked according to their importance, support, and confidence and selected in sequence to build a classifier with high accuracy. An example illustrates how this approach works. [ABSTRACT FROM AUTHOR]

In the paper, two families of lazy classification algorithms of polynomial time complexity are considered. These algorithms are based on ordinary and inhibitory rules, but the direct generation of rules is not required. Instead of this, the considered algorithms extract efficiently for a new object some information on the set of rules which is next used by a decision-making procedure. [ABSTRACT FROM AUTHOR]

This paper presents the results of our experiments on a data set describing neonatal infection. We used two main tools: the MLEM2 algorithm of rule induction and BeliefSEEKER system for generation of Bayesian nets and rule sets. Both systems are based on rough set theory. Our main objective was to compare the quality of diagnosis of cases from two testing data sets: with an additional attribute called PCT and without this attribute. The PCT attribute was computed using constructive induction. The best results were associated with the rule set induced by the MLEM2 algorithm and testing data set enhanced by constructive induction. [ABSTRACT FROM AUTHOR]

A framework of Non-deterministicInformationSystems (NISs) is known well for handling information incompleteness in Deter- ministicInformationSystems (DISs). Apriori algorithm for the standard tables or DISs is also known as an algorithm to generate rules, which are characterized by criteria, support and accuracy. This paper extends Apriori algorithm in DISs to Apriori algorithm in NISs. This extended Apriori algorithm employs criteria, minimumsupport and minimumaccuracy in NISs, and generates rules under the worst condition. A software tool is also implemented. [ABSTRACT FROM AUTHOR]

In this paper, a decision table in rough set theory is classified into three types according to its consistency. Three parameters α (whole certainty measure), β (whole consistency measure) and γ (whole support measure) are introduced to evaluate the performance of a decision rule set induced from a decision table. For three types of decision tables, the dependency of the parameters upon condition/decision granulation is analyzed. The parameters can be used to construct an evaluation function in favor of selecting a better one from some different rule acquiring methods for real decision problems. [ABSTRACT FROM AUTHOR]

Constructing accurate classifier based on association rule is an important and challenging task in data mining. In this paper, a novel combination strategy based on rough sets (RST) and evidence theory (DST) for associative classification (RSETAC) is proposed. In RSETAC, rules are regarded as classification experts, after the calculation of the basic probability assignments (bpa) according to rule confidences and evidence weights employing RST, Yang's rule of combination is employed to combine the distinct evidences to realize an aggregate classification. A numerical example is shown to highlight the procedure of the proposed method. The comparison with popular methods like CBA, C4.5, RIPPER and MCAR indicates that RSETAC is a competitive method for classification based on association rule. [ABSTRACT FROM AUTHOR]

The Web contains massive amount of documents to the point where it has become impossible to classify them manually. This project's goal is to find a new method for clustering documents that is as close to humans' classification as possible and at the same time to reduce the size of the documents. This project uses a combination of Latent Semantic Indexing (LSI) with Singular Value Decomposition (SVD) calculation and Support Vector Machine (SVM) classification. Using SVD, data is decomposed and truncated to reduce the data size. The reduced data will be clustered into different categories. Using SVM, clustered data from SVD calculation is used for training to allow new data to be classified based on SVM's prediction. The project's result show that the method of combining SVD and SVM is able to reduce data size and classifies documents reasonably compared to humans' classification. [ABSTRACT FROM AUTHOR]

Suffix trees and suffix arrays are two well-known index data structures for strings. It is known that the latter can be easily transformed into the former: Iliopoulos and Rytter [5] showed two simple transformation algorithms on the CREW PRAM model. However, the PRAM model is a theoretical one and we need a practical parallel model. The Message Passing Interface (MPI) is a standard widely used on both massively parallel machines and on clusters. In this paper, we show how to implement the algorithms of Iliopoulos and Rytter on the MPI environment. Our contribution includes the modification of algorithms due to the lack of shared memory, small number of processors, communication costs between processors. [ABSTRACT FROM AUTHOR]

When samples number, classification and dimension of clustering are much more, traditional clustering algorithm usually leads to unharmonious character between clustering and transcendent knowledge. Therefore, a new clustering algorithm is proposed, which is parallel artificial immune clustering algorithm based on granular computing. Artificial immune system model has the characteristics, such as parallel, random searching and maintaining diversity, which can solve premature problem in latter evolution and converge to a global optimization solution faster. Besides, we unite it to dynamic granulation model and apply granulation description to clustering. In the process of granulation changing, we can choose appropriate granulation size by adjusting to ensure clustering efficiency and quality. Tests show that the algorithm is more effective and more reasonable when we handle clustering of some data with it. [ABSTRACT FROM AUTHOR]

Reading and literature exploration are important tasks of scientific research. However, conventional retrieval systems provide limited support for these tasks by concentrating on identifying relevant materials. New generation systems should provide additional support functionality by focusing on analyzing and organizing the retrieved materials. A framework of literature exploration support systems is proposed. Techniques of granular computing are used to construct granular knowledge structures from the contents, structures, and usages of scientific documents. The granular knowledge structures provide a high level understanding of scientific literature and hints regarding what has been done and what needs to be done. As a demonstration, we examine granular knowledge structures obtained from an analysis of papers from two rough sets related conferences. [ABSTRACT FROM AUTHOR]

The Data Credibility Analysis is a computer science domain aimed at discovering universal algorithms for identifying improper or unusual data. It is done by calculating credibility coefficients for individual records. In recent years many different methods of computing these coefficients were presented. In the paper we propose a transformation of credibility coefficients to ordinal credibility coefficients. By developing this idea we propose another credibility coefficient computing algorithm, which benefits from incorporating arbit rary many other credibility coefficient computing methods. The preliminary tests showed that this approach leads to better results. [ABSTRACT FROM AUTHOR]

Multiscale extraction of the subtlest signs of hypodensity, which were often undetected in standard CT scan review was the subject of our research. Proposed method is as follows: evidence-based description of hypodense changes, the investigation of hypodensity across scales, basing on a set of over 20 hyperacute stroke exams, the improvement of wavelet-based display of ischemic stroke. Considered problems were: -extension of the brain tissues for marginal and missing space after deskulling and segmenting of unusual areas; -best basis selection;-non-perfect reconstruction across scales as an extraction of hypo-attenuating tendency. Increased visibility of hypodense signs on CT scans performed in patients with hyperacute stroke was noticed in subjective rating. In opinion of radiologists and image processing experts, enhanced perception of hypodense area was noticed for all test exams. The rate of unique, clear and doubtless extraction of hypodense area was 92% in 13 tested cases of hyperacute ischemic stroke. [ABSTRACT FROM AUTHOR]

Rough set theory has become well-established as a mechanism for uncertainty management in a wide variety of applications. This paper studies the measurement of uncertainty in generalized fuzzy rough sets determined by a triangular norm. Based on information theory, the entropy of a generalized fuzzy approximation space is introduced, which is similar to Shannon's entropy. To measure uncertainty in generalized fuzzy rough sets, a notion of fuzziness is introduced. Some basic properties of this measure are examined. For a special triangular norm T =  min , it is proved that the measure of fuzziness of a generalized fuzzy rough set is equal to zero if and only if the set is crisp and definable. [ABSTRACT FROM AUTHOR]

In this paper we present FuzzyPR, a novel fuzzy logic based passage retrieval system for Question Answering Systems (QAS). FuzzyPR employs a fuzzy logic based similarity measure that includes the best performing models to implement the question reformulation intuition. Our experiments show that FuzzyPR achieves consistently better performance in terms of coverage than JIRS on the TREC corpora and slightly better on the CLEF corpora. [ABSTRACT FROM AUTHOR]

This paper demonstrates a study on lateral ventricles segmentation in brain Magnetic Resonance Imaging (MRI). The method applies Gaussian smoothed image data as additional features into the feature space of Fuzzy C-Means (FCM) algorithm. With the aid of the smoothing effect from Gaussian filters, FCM is able to segment lateral ventricular compartments by reducing inappropriate clustering caused by noise and inhomogeneous intensity distribution. The results demonstrate both noise insensitivity and more homogeneous clustering. [ABSTRACT FROM AUTHOR]

Knowledge discovery methods used to find relationships among software engineering data and the extraction of rules have gained increasing importance in recent years. These methods have become necessary for improvements in the quality of the software product and the process. The focus of this paper is a first attempt towards combining strengths of rough set theory and neuro-fuzzy decision trees in classifying software defect data. We compare classification results for four methods: rough sets, neuro-fuzzy decision trees, partial decision trees, rough-neuro-fuzzy decision trees. The analysis of the results include a family-wise 10 fold paired t-test for accuracy and number of rules. The contribution of this paper is the application of a hybrid rough-neuro-fuzzy decision tree method in classifying software defect data. [ABSTRACT FROM AUTHOR]

This paper discusses the problems arising in applications of the unified rough and fuzzy rough set approach to analysis of inconsistent information systems. The unified approach constitutes a parameterized generalization of the variable precision rough set model. It bases on a single notion of parameterized ε-approximation. As a necessary extension, a method suitable for a correct determination of attributes' significance is proposed. In particular, the notions of positive ε-classification region and ε-approximation quality are considered. A criterion for reduction of condition attributes is given. Furthermore, a generalized definition of the fuzzy extension ω is proposed. [ABSTRACT FROM AUTHOR]

This paper presents a general framework for the study of covering-based fuzzy approximation operators in which a fuzzy set can be approximated by some elements in a crisp or a fuzzy covering of the universe of discourse. Two types of approximation operators, crisp-covering-based rough fuzzy approximation operators and fuzzy-covering-based fuzzy rough approximation operators, are defined, their properties are examined in detail. Finally, the comparison of these new approximation operators is done, a sufficient and necessary condition is given under which some operators are equivalent, and approximation operator characterization of fuzzy partitions of the universe is obtained. [ABSTRACT FROM AUTHOR]

The classification of the background and cell areas is very important but difficult problem due to the ambiguity of boundaries. In this paper, the cell region is extracted from an image of uterine cervical cytodiagnosis using the region growing method. Segmented images from background and cell areas are binarized using a threshold value. And the 8-directional tracking algorithm for contour lines is applied to extract the cell area. Each extracted nucleus is transformed to the original RGB space. Then the K-Means clustering algorithm is employed to classify RGB pixels to the R, G, and B channels, respectively. Finally, the Hue information of nucleus is extracted from the HSI models that are transformed using the clustering values in R, G, and B channels. The fuzzy RBF Network is then applied to classify and identify the normal or abnormal nucleus. The result shows that the accuracy of our method is 80% overall and 66% in 5-class problem according to the Bethesda system. [ABSTRACT FROM AUTHOR]

We propose a method to classify breast lesions of ducatal origin. The materials were tissue sections of the intraductal proliferative lesions of the breast: benign(DH:ductal hyperplasia), ductal carcinoma in situ(DCIS). The total 40 images from 70 samples of ducts were digitally captured from 15 cases of DCIS and 25 cases of DH diagnosed by pathologist. To assess the correlation between computerized images analysis and visual analysis by a pathologist, we extracted the total lumen area/gland area, to segment the gland(duct) area used a snake algorithm, to segment the lumen used multilevel Otsus method in the duct from 20x images for distinguishing DH and DCIS. In duct image, we extracted the five texture features (correlation, entropy, contrast, angular second moment, and inverse difference moment) using the co-occurrence matrix for a distribution pattern of cells and pleomorphism of the nucleus. In the present study, we obtained classification accuracy rates of 91.33%, the architectural features of breast ducts has been advanced as a useful features in the classification for distiguishing DH and DCIS. We expect that the proposed method in this paper could be used as a useful diagnostic tool to differentiate the intraductal proliferative lesions of the breast. [ABSTRACT FROM AUTHOR]

This paper presents a new fuzzy classifier design, which constructs one classifier for each fuzzy partition of a given system. The new approach, namely Fuzzy Classifier Functions (FCF), is an adaptation of our generic design on Fuzzy Functions to classification problems. This approach couples any fuzzy clustering algorithm with any classification method, in a unique way. The presented model derives fuzzy functions (rules) from data to classify patterns into number of classes. Fuzzy c-means clustering is used to capture hidden fuzzy patterns and a linear or a non-linear classifier function is used to build one classifier model for each pattern identified. The performance of each classifier is enhanced by using corresponding membership values of the data vectors as additional input variables. FCF is proposed as an alternate representation and reasoning schema to fuzzy rule base classifiers. The proposed method is evaluated by the comparison of experiments with the standard classifier methods using cross validation on test patterns. [ABSTRACT FROM AUTHOR]