Interpretable knowledge discovery from data with DC*

Information Sciences, 2008

Information granules are complex entities that arise in the process of abstraction of data and derivation of knowledge. The automatic generation of information granules from data is an important task, since it gives to machines the ability of acquiring knowledge that can be communicated to users. For this purpose, knowledge acquisition should provide for fuzzy information granules that can be naturally labelled by linguistic terms, i.e. symbols that belong to the natural language. Information granules of this type are called interpretable. However, interpretability cannot be guaranteed until a set of constraints is imposed on the granulation process. In literature, several interpretability constraints have been proposed, but, due to the subjective interpretation of interpretability, there is no agreement on which constraints should be adopted. This survey is an attempt to provide for a complete presentation of interpretability constraints adopted in literature with the following objectives: (i) to give a homogeneous description of all interpretability constraints; (ii) to provide for a critical review of such constraints; (iii) to identify potentially different meanings of interpretability. Hopefully, this survey may serve as a guidance for designing interpretable fuzzy models as well as for identifying new methods of interpretable information granulation.

2005

Abstract Granular Computing is an emerging conceptual and computational paradigm for information processing, which concerns representation and processing of complex information entities called “information granules” arising from processes of data abstraction and knowledge derivation. Within Granular Computing, a prominent position is assumed by the “Theory of Fuzzy Information Granulation”(TFIG) whose centrality is motivated by the ability of representing and processing perception-based granular information.

2012

Abstract Clustering forms one of the most visible conceptual and algorithmic framework of developing information granules. In spite of the algorithm being used, the representation of information granules-clusters is predominantly numeric (coming in the form of prototypes, partition matrices, dendrograms, etc.). In this paper, we consider a concept of granular prototypes that generalizes the numeric representation of the clusters and, in this way, helps capture more details about the data structure.

Proceedings of the 2015 Conference of the International Fuzzy Systems Association and the European Society for Fuzzy Logic and Technology, 2015

Rule-level feature selection, also termed as rule compression, is an important technique for improving interpretability of fuzzy rule-based classifiers. In this paper we present three different rule compression algorithms and analyze their performance and characteristics on the classifiers identified from wellknown classification benchmarks, namely the Iris, Wine and two versions of Wisconsin Breast Cancer data sets. Our study shows that the classifiers, in which the overlap between either the rules representing different classes or all rules is eliminated, can be usually compressed at a higher rate and that the interpretation of such classifiers is more insightful.

Proceedings of the International Joint Conference on Neural Networks, 2003.

Cluster Analysis is traditionally an unsupervised data reduction technique. However, hy unifying ideas from both Cluster Analysis and Rough Set Theory, the inherent structure of a data set can be expressed as a rule set which, in turn, can be modified in a supervised manner to obtain a decision .rule set with minimal ambiguity. This process of encasing knowledge extraction in an algorithmic framework results in an optimal process for decision support.

Mining association rules is a common data mining technique to find knowledge about the universe [3]. Due to the lack of sufficient information about the universe we can not uniquely identify each element of the universe. To characterize the elements of the universe and to extract knowledge about the universe we classify the elements of the universe based on indiscernibility relation. The classification results in a set of classes called granules which are basic building blocks of the knowledge about the universe [3, 4]. Granular computing processes these granules and produces possible patterns and associations. In this paper we study the concept of indiscernibility and knowledge granulation. We study the association rules using conventional method and granular computing method. We process the granules produced as the result of classification and produce association rules using granular computing. Finally, we compare both the methods in terms of finding association rules.

2002

Abstract This paper proposes an approach to derive fuzzy granules from numerical data. Granules are first formed by means of a doubleclustering technique, and then properly fuzzified so as to obtain interpretable granules, in the sense that they can be described by linguistic labels. The double-clustering technique involves two steps. First, information granules are induced in the space of numerical data via the FCM algorithm.

IEEE Transactions on Information Technology in Biomedicine, 2007

Clinical prediction rules play an important role in medical practice. They expedite diagnosis and limit unnecessary tests. However, the rule creation process is time consuming and expensive. With the current developments of efficient data mining algorithms and growing accessibility to medical data, the creation of clinical rules can be supported by automated rule induction from data. A data-driven method based on the reuse of previously collected medical records and clinical trial statistics is cost-effective; however, it requires well defined and intelligent methods for data analysis. This paper presents a new framework for knowledge representation for secondary data analysis and for generation of a new typicality measure, which integrates medical knowledge into statistical analysis. The framework is based on a semiotic approach for contextual knowledge and fuzzy logic for approximate knowledge. This semio-fuzzy framework has been applied to the analysis of predictors for the diagnosis of obstructive sleep apnea. This approach was tested on two clinical data sets. Medical knowledge was represented by a set of facts and fuzzy rules, and used to perform statistical analysis. Statistical methods provided several candidate outliers. Our new typicality measure identified those, which were medically significant, in the sense that the removal of those important outliers improved the descriptive model. This is a critical preprocessing step towards automated induction of predictive rules from data. These experimental results demonstrate that knowledge-based methods integrated with statistical approaches provide a practical framework to support the generation of clinical prediction rules.

2010

More and more, the analysis of clustering results becomes difficult as the number of variables considered increases, and the number of classes is not low. Sometimes concept induction methods are used to associate concepts to every class and use to be expressed as boolean expressions, easy to understand and supposedly providing good support to decision making. It has been seen that most of the concept induction algorithms priorize the compacity of the final expressions, as well as their predictive power. However, for descriptive purposes, when the meaning of classes has to be recognized and understood by the expert, this is not the best approach, since compacity directly implies elimination of redundancies or strong associations, while comprehension of the class is mainly based in understanding how variables interact among them inside the class. Here a method to induce conceptual descriptions of classes is proposed, providing non-minimal descriptions of the classes, but richer ones i...

2002

In supervised learning the inductive algorithm seeks to develop a conceptual description, or prescriptive model, from examples or objects that have been pre-classified. On the other hand, in unsupervised learning, or clustering, the task of the algorithm is to group non-classified examples into clusters of examples, building a descriptive model. Central to all of the goals of clustering is the notion of similarity (or dissimilarity) between the individual objects being clustering. The key distinction between prediction and description is that prediction has as its objective a unique variable, known as the class attribute in supervised learning, while in descriptive problems no single variable is central to the problem. In this work we propose an approach, as well as a particular implementation of this approach, were a combination of traditional clustering and inductive learning are used to find and interpret special patterns from a database in order to extract useful knowledge to represent real world domains. This approach helps the user to make explicit his/her prior knowledge relevant to the problem allowing to organize the set of objects in different clusters into a set of classes considered to be semantically interpretable by the user.