## Data Mining with Fuzzy Methods: Status and Perspectives (1999)

### Cached

### Download Links

Venue: | In: Proceedings of the EUFIT’99 |

Citations: | 3 - 0 self |

### BibTeX

@INPROCEEDINGS{Kruse99datamining,

author = {Rudolf Kruse and Detlef Nauck and Christian Borgelt},

title = {Data Mining with Fuzzy Methods: Status and Perspectives},

booktitle = {In: Proceedings of the EUFIT’99},

year = {1999}

}

### OpenURL

### Abstract

: Data mining is the central step in a process called knowledge discovery in databases, namely the step in which modeling techniques are applied. Several research areas like statistics, artificial intelligence, machine learning, and soft computing have contributed to its arsenal of methods. In this paper, however, we focus on fuzzy methods for rule learning, information fusion, and dependency analysis. In our opinion fuzzy approaches can play an important role in data mining, because they provide comprehensible results (although this goal is often neglected---maybe because it is sometimes hard to achieve with other methods). In addition, the approaches studied in data mining have mainly been oriented at highly structured and precise data. However, we expect that the analysis of more complex heterogeneous information source like texts, images, rule bases etc. will become more important in the near future. Therefore we give an outlook on information mining, which we see as an extension o...

### Citations

7441 |
Probabilistic Reasoning in Intelligent Systems
- Pearl
- 1988
(Show Context)
Citation Context ...e and edge processors in the conditional independence graph. Using graphical models to facilitate reasoning in multi-dimensional domains has originated in the probabilistic setting. Bayesian networks =-=[30]-=-, which are based on directed conditional independence graphs, and Markov networks [26], which are based on undirected graphs, are the most prominent examples. However, this approach has also been gen... |

1343 |
Local Computations with Probabilities on Graphical Structures and their Application to Expert Systems
- Lauritzen, Spiegelhalter
- 1988
(Show Context)
Citation Context ...facilitate reasoning in multi-dimensional domains has originated in the probabilistic setting. Bayesian networks [30], which are based on directed conditional independence graphs, and Markov networks =-=[26]-=-, which are based on undirected graphs, are the most prominent examples. However, this approach has also been generalized to be usable with other uncertainty calculi [32], for instance, in the so-call... |

1132 | A Bayesian method for the induction of probabilistic networks from data
- Cooper, Herskovits
- 1992
(Show Context)
Citation Context ...ample cases. Due to the origin of graphical modeling research in probabilistic reasoning (see above), the most widely known methods are, of course, learning algorithms for Bayesian or Markov networks =-=[7, 8, 20]-=-. However, these approaches---as probabilistic approaches do in general---suffer from certain deficiencies, if imprecise information, understood as set-valued data, has to be taken into account. For t... |

949 | Learning Bayesian networks: The combination of knowledge and statistical data
- Heckerman, Geiger, et al.
- 1995
(Show Context)
Citation Context ...ample cases. Due to the origin of graphical modeling research in probabilistic reasoning (see above), the most widely known methods are, of course, learning algorithms for Bayesian or Markov networks =-=[7, 8, 20]-=-. However, these approaches---as probabilistic approaches do in general---suffer from certain deficiencies, if imprecise information, understood as set-valued data, has to be taken into account. For t... |

682 | Approximating discrete probability distributions with dependence trees
- Chow, Liu
- 1968
(Show Context)
Citation Context ...ample cases. Due to the origin of graphical modeling research in probabilistic reasoning (see above), the most widely known methods are, of course, learning algorithms for Bayesian or Markov networks =-=[7, 8, 20]-=-. However, these approaches---as probabilistic approaches do in general---suffer from certain deficiencies, if imprecise information, understood as set-valued data, has to be taken into account. For t... |

465 |
Graphical models in applied multivariate statistics
- Whittaker
- 1990
(Show Context)
Citation Context ...and imprecision are involved---decomposition techniques, that reduce the reasoning process to computations in lower-dimensional subspaces, have become very popular. In the field of graphical modeling =-=[23, 36]-=-, decomposition is based on dependence and independence relations between the attributes or variables that are used to describe the domain under consideration. The structure of these dependence and in... |

363 |
Generating fuzzy rules by learning from examples
- Wang, Mendel
- 1992
(Show Context)
Citation Context ...onal fuzzy grid. A rule base is created by selecting those grid cells that contain data. This can be done in a single pass through the training data. This way of learning fuzzy rules was suggested in =-=[35]-=-. Extended versions were used in the the neuro-fuzzy classification system NEFCLASS [28]. NEFCLASS uses a performances measure for the detected fuzzy rules. Thus the size of the rule base can be deter... |

157 | Hugin: a shell for building bayesian belief universes for expert systems - Andreassen, Olesen, et al. - 1989 |

152 |
Fundations of neuro-fuzzy systems
- Nauck, Klawonn, et al.
- 1997
(Show Context)
Citation Context ... a rule base (structure) and fuzzy partitions (parameters) for all variables. The data driven induction of fuzzy systems by simple heuristics based on local computations is usually called neuro-fuzzy =-=[28]-=-. If we apply such techniques, we must be aware of the trade-off between precision and interpretability. A solution is not only judged by its accuracy, but also---if not primarily---by its simplicity ... |

144 | Probabilistic Similarity Networks - Heckerman - 1990 |

144 |
Foundations of Fuzzy Systems
- Kruse, Gebhardt, et al.
- 1994
(Show Context)
Citation Context ..., fuzzy set theory makes it possible that an object or a case belongs to a set only to a certain degree, thus modeling the penumbra of the linguistic term describing the property that defines the set =-=[24]-=-. Interpretations of membership degrees include similarity, preference, and uncertainty [11]: They can state how similar an object or case is to a prototypical one, they can indicate preferences betwe... |

136 |
Fuzzy Cluster Analysis
- Höppner, Klawonn, et al.
- 1999
(Show Context)
Citation Context ...ed, transformed and maybe properly scaled to produce the input for the modeling techniques. In this step fuzzy methods may, for example, be used to detect outliers, e.g., by fuzzy clustering the data =-=[3, 21]-=- and then finding those data points that are far away from the cluster prototypes. The modeling phase, in which models are constructed from the data in order, for instance, to predict future developme... |

121 |
Fuzzy models and algorithms for pattern recognition and image processing
- Bezdek, Keller, et al.
- 1999
(Show Context)
Citation Context ...ed, transformed and maybe properly scaled to produce the input for the modeling techniques. In this step fuzzy methods may, for example, be used to detect outliers, e.g., by fuzzy clustering the data =-=[3, 21]-=- and then finding those data points that are far away from the cluster prototypes. The modeling phase, in which models are constructed from the data in order, for instance, to predict future developme... |

66 |
Valuation-based systems: A framework for managing uncertainty in expert systems. Pages 83–104
- Shenoy
- 1992
(Show Context)
Citation Context ...ted graphs, are the most prominent examples. However, this approach has also been generalized to be usable with other uncertainty calculi [32], for instance, in the so-called valuation-based networks =-=[33]-=- and has been implemented, for example, in PULCINELLA [31]. Due to their connection to fuzzy systems, which in the past have successfully been applied to solve control problems, and due to their abili... |

63 |
Experts in uncertainty
- Cooke
- 1991
(Show Context)
Citation Context ...r levels, where raw data is involved, the term (sensor) data fusion is preferred. For a conceptual and comparative study of fusion strategies for expert opinions in various calculi of uncertainty see =-=[10, 12, 17]-=-. NEFCLASS and neuro-fuzzy systems in general can be used to integrate expert knowledge in form of fuzzy rules and information obtained from data. If prior knowledge about the classification problem i... |

47 | PULCINELLA: A General Tool for Propagating Uncertainty in Valuation Networks
- Saffiotti, Umkehrer
- 1991
(Show Context)
Citation Context ... approach has also been generalized to be usable with other uncertainty calculi [32], for instance, in the so-called valuation-based networks [33] and has been implemented, for example, in PULCINELLA =-=[31]-=-. Due to their connection to fuzzy systems, which in the past have successfully been applied to solve control problems, and due to their ability to deal not only with uncertainty but also with impreci... |

26 |
The context model an integrating view of vagueness and uncertainty Int
- Gebhardt, Kruse
- 1993
(Show Context)
Citation Context ...y but also with imprecision, recently possibilistic networks also gained some attention. They can be based on the context-model interpretation of a degree of possibility, which focuses on imprecision =-=[14]-=-, and have been implemented, for example, in POSSINFER [16, 24]. For some time the standard approach to construct a graphical model has been to let a human domain expert specify the dependency structu... |

25 | Learning possibilistic networks from data
- Gebhardt, Kruse
- 1995
(Show Context)
Citation Context ...a, has to be taken into account. For this reason learning possibilistic networks from data is a noteworthy alternative, the theory of which can be developed in close analogy to the probabilistic case =-=[15, 16, 4, 5, 22]-=-. These methods can be used to do dependency analysis, even if the data to analyze is highly imprecise and thus offer interesting perspectives for future research. 5 Perspective: Information Mining Al... |

22 | Evaluation measures for learning probabilistic and possibilistic networks
- Borgelt, Kruse
- 1997
(Show Context)
Citation Context ...a, has to be taken into account. For this reason learning possibilistic networks from data is a noteworthy alternative, the theory of which can be developed in close analogy to the probabilistic case =-=[15, 16, 4, 5, 22]-=-. These methods can be used to do dependency analysis, even if the data to analyze is highly imprecise and thus offer interesting perspectives for future research. 5 Perspective: Information Mining Al... |

19 |
Tightest hypertree decompositions of multivariate possibility distributions
- Gebhardt, Kruse
- 1996
(Show Context)
Citation Context ...s also gained some attention. They can be based on the context-model interpretation of a degree of possibility, which focuses on imprecision [14], and have been implemented, for example, in POSSINFER =-=[16, 24]-=-. For some time the standard approach to construct a graphical model has been to let a human domain expert specify the dependency structure of the considered domain. This provided the conditional inde... |

19 |
Local Computations in Hypertrees (Working Paper 201
- Shafer, Shenoy
- 1988
(Show Context)
Citation Context ...e graphs, and Markov networks [26], which are based on undirected graphs, are the most prominent examples. However, this approach has also been generalized to be usable with other uncertainty calculi =-=[32]-=-, for instance, in the so-called valuation-based networks [33] and has been implemented, for example, in PULCINELLA [31]. Due to their connection to fuzzy systems, which in the past have successfully ... |

18 | Fuzzy Data Analysis: Challenges and Perspectives
- Kruse, Bolgelt, et al.
- 1999
(Show Context)
Citation Context ...s obtained using fuzzy approaches are easy to understand and to apply. Due to these strengths, fuzzy systems are the method of choice, if linguistic, vague, or imprecise information has to be modeled =-=[25]-=-. Data mining is concerned with the analysis of large but homogeneous data. Our information-oriented world, however, demands that we also automate the analysis of more complex information sources like... |

16 |
Merging fuzzy information
- Dubois, Prade, et al.
- 1999
(Show Context)
Citation Context ...r levels, where raw data is involved, the term (sensor) data fusion is preferred. For a conceptual and comparative study of fusion strategies for expert opinions in various calculi of uncertainty see =-=[10, 12, 17]-=-. NEFCLASS and neuro-fuzzy systems in general can be used to integrate expert knowledge in form of fuzzy rules and information obtained from data. If prior knowledge about the classification problem i... |

14 |
editors. Uncertainty and vagueness
- Kruse, Schwecke, et al.
- 1991
(Show Context)
Citation Context ...and imprecision are involved---decomposition techniques, that reduce the reasoning process to computations in lower-dimensional subspaces, have become very popular. In the field of graphical modeling =-=[23, 36]-=-, decomposition is based on dependence and independence relations between the attributes or variables that are used to describe the domain under consideration. The structure of these dependence and in... |

10 | Constructing fuzzy graphs from examples
- Berthold, Huber
- 1999
(Show Context)
Citation Context ... cluster analysis [3, 21], i.e., the learning process is unsupervised. Hyperbox-oriented approaches use a supervised learning algorithm that tries to cover the training data by overlapping hyperboxes =-=[2]-=-. Fuzzy rules are created in both approaches by projection of clusters or hyperboxes. The main problem of both approaches is that each generated fuzzy rule uses individual membership functions and thu... |

7 | Data Mining with Graphical Models
- Borgelt
- 2000
(Show Context)
Citation Context ...a, has to be taken into account. For this reason learning possibilistic networks from data is a noteworthy alternative, the theory of which can be developed in close analogy to the probabilistic case =-=[15, 16, 4, 5, 22]-=-. These methods can be used to do dependency analysis, even if the data to analyze is highly imprecise and thus offer interesting perspectives for future research. 5 Perspective: Information Mining Al... |

5 |
Chapter F1.2: Inference Methods
- Borgelt, Gebhardt, et al.
- 1998
(Show Context)
Citation Context |

5 |
Parallel combination of information sources
- Gebhardt
- 1998
(Show Context)
Citation Context ...r levels, where raw data is involved, the term (sensor) data fusion is preferred. For a conceptual and comparative study of fusion strategies for expert opinions in various calculi of uncertainty see =-=[10, 12, 17]-=-. NEFCLASS and neuro-fuzzy systems in general can be used to integrate expert knowledge in form of fuzzy rules and information obtained from data. If prior knowledge about the classification problem i... |

4 |
Wissensentdeckung in Datenbanken und Data Mining
- Nakhaeizadeh, Reinartz, et al.
- 1998
(Show Context)
Citation Context ... the KDD process. It is an interdisciplinary field that employs methods from statistics, soft computing, artificial intelligence and machine learning. Usually data mining is defined by a set of tasks =-=[13, 27]-=-, which include at least segmentation (e.g. what kind of customers does a company have?), classification (e.g. is this person a prospective customer?), concept description (e.g. what attributes descri... |

4 |
Chapter D.2: Neurofuzzy Systems
- Nauck, Kruse
- 1998
(Show Context)
Citation Context ...luster-oriented and hyperbox-oriented approaches to fuzzy rule learning create rules and fuzzy sets at the same time. Structure-oriented approaches need initial fuzzy partitions to create a rule base =-=[29]-=-. Cluster-oriented rule learning approaches are based on fuzzy cluster analysis [3, 21], i.e., the learning process is unsupervised. Hyperbox-oriented approaches use a supervised learning algorithm th... |

4 |
Information Fusion in the Context of Stock Index Prediction
- Siekmann, Kruse, et al.
- 1999
(Show Context)
Citation Context ...rules by descending performances values. For an application in the context of stock prediction that is based on another neuro-fuzzy approach, which is implemented in the neural network tool SENN, see =-=[34]-=-. The decision on that option depends on the trust we have in the experts knowledge and in the training data. A mixed approach can be used, e.g. include the best expert rules and then use the best lea... |

3 |
Information Engineering and Fuzzy Logic
- Dubois, Prade, et al.
- 1996
(Show Context)
Citation Context ...tain degree, thus modeling the penumbra of the linguistic term describing the property that defines the set [24]. Interpretations of membership degrees include similarity, preference, and uncertainty =-=[11]-=-: They can state how similar an object or case is to a prototypical one, they can indicate preferences between suboptimal solutions to a problem, or they can model uncertainty about the true situation... |

1 |
Fuzzy Methoden in der Datenanalyse
- Borgelt, Gebhardt, et al.
- 1999
(Show Context)
Citation Context ...wo classes. The first class, fuzzy data analysis, consists of approaches that analyze fuzzy data---data derived from imprecise measurement instruments or from the descriptions of human domain experts =-=[6]-=-. An example from our own research is the induction of possibilistic graphical models from data which complements the induction of the well-known probabilistic graphical models. The second class, fuzz... |

1 |
Available from http://www.ncr.dk/CRISP
- Chapman, Clinton, et al.
- 1999
(Show Context)
Citation Context ... which can be expected to have considerable impact, since it is backed by several large companies like NCR and DaimlerChrysler, is the CRISP-DM model (CRoss Industry Standard Process for Data Mining) =-=[9]-=-. The basic structure of this process model is depicted in figure 1. The circle indicates that data mining is essentially a circular process, in which the evaluation of the results can trigger a re-ex... |

1 |
Data Mining Tools: Vergleich marktgangiger Tools
- Gentsch
- 1999
(Show Context)
Citation Context ...g, and association rules. In this paper we concentrate on fuzzy methods in data mining and show where and how they can be used. A survey of commercial data mining tools can be found, for instance, in =-=[18]. Fuzzy se-=-t theory provides excellent means to model the "fuzzy" boundaries of linguistic terms by introducing gradual memberships. In contrast to classical set theory, in which an object or a case ei... |

1 |
Data Mining with Graphical Models. Proc. Computer Science for Environmental Protection (12th Int. Symp. "Informatik fr den Umweltschutz
- Kruse, Borgelt
- 1998
(Show Context)
Citation Context |