Anna Stachowiak

An Interval-Valued Fuzzy Classifier Based on an Uncertainty-Aware Similarity Measure

In this paper we propose a new method for classifying uncertain data, modeled as interval-valued fuzzy sets. We develop the notion of an interval-valued prototype-based fuzzy classifier, with the idea of preserving full information including the uncertainty factor about data during the classification process. To this end, the classifier was based on the uncertainty-aware similarity measure, a new concept which we introduce and give an axiomatic definition for. Moreover, an algorithm for determining such a similarity value is proposed, and an application to supporting medical diagnosis is described.

A Bipolar View on Medical Diagnosis in OvaExpert System

In the paper we present OvaExpert - a unique tool for supporting gynecologists in the diagnosis of ovarian tumor, combining classical diagnostic scales with modern methods of machine learning and soft computing. A distinguishing feature of the system is its comprehensiveness, which makes it usable at any stage of a diagnostic process. We gather all the results and solutions making up the system, some of which were described in our other publications, to provide an overall picture of OvaExpert and its capabilities. A special attention is paid to a property of supporting uncertainty modeling and processing, that is an essential part of the system.

External validation of the IOTA ADNEX model performed by two independent gynecologic centers

OBJECTIVES The external, two-center validation of the IOTA ADNEX model for differential diagnosis of adnexal tumors. METHODS A total of 204 patients with adnexal masses (134 benign and 70 malignant) treated at the Division of Gynecologic Surgery, Poznan University of Medical Sciences, Poland (Center I), and 123 patients (89 benign and 34 malignant) from the Department of Obstetrics and Gynecology, Clinica Universidad de Navarra, University of Navarra School of Medicine, Pamplona, Spain (Center II), were enrolled into the study. RESULTS ADNEX achieved high accuracy in discriminating between malignant and benign ovarian tumors in both centers (79.9% and 81.3% in Centers I and II, respectively). Multiclass accuracy was substantially lower than in binary classification (malignant vs. benign): 64.2% and 74.0% in Centers I and II, respectively. Sensitivity and specificity for the diagnosis of specific tumor types in Center I were as follows: benign tumors - 72.4% and 94.3%; borderline tumors - 33.3% and 87.0%, stage I ovarian cancers - 00.0% and 91.8%; stage II-IV ovarian cancers - 68.2% and 83.1%; and metastatic tumors - 00.0% and 99.5%. Sensitivity and specificity in Center II were as follows: benign tumors - 75.3% and 97.1%; borderline tumors - 50.0% and 88.2%, stage I ovarian cancers - 40.0% and 97.5%; stage II-IV ovarian cancers - 95.0% and 88.3%; and metastatic tumors - 20.0% and 98.3%. CONCLUSIONS ADNEX is characterized by very high accuracy in differentiating between malignant and benign adnexal tumors. However, prediction of ovarian tumor types could be more accurate.

An algorithmic study of relative cardinalities for interval-valued fuzzy sets

The main topic of this paper is the notion of relative cardinality for interval-valued fuzzy sets its definition, properties and computation. First we define relative cardinality for interval-valued fuzzy sets following the concept of uncertainty modelling given by Mendels Wavy-Slice Representation Theorem. We expand on previous approaches by considering relative cardinality based on different t-norms and scalar cardinalities and we initiate an investigation of its properties and possible applications. Drawing on the NguyenKreinovich and KarnikMendel algorithms, we propose efficient algorithms to compute relative cardinality depending on a chosen t-norm. This seems to be the first such broad and consistent analysis to have been made of relative cardinality for interval-valued fuzzy sets. As a promising application we consider using interval-valued relative cardinality to construct the family of parameterised subsethood measures. We define generalised relative cardinality for IVFS using different t-norms.We study properties of such relative cardinality especially as a subsethood measure.We construct efficient algorithms to compute such relative cardinality for IVFS.We discuss practical implementation and applications of proposed algorithms.

An Intelligent System for Computer-Aided Ovarian Tumor Diagnosis

This article describes the fundamentals of an intelligent decision support system for the diagnosis of ovarian tumors. The system is designed to support diagnosis by less experienced gynecologists, and to gather data for continuous improvement of the quality of diagnosis. The theoretical basis for the construction of the system is the IF-sets framework, used to aggregate multiple decision-making methods, and simultaneously providing information about positive and negative diagnosis of a given tumor type.

A Recommender System with Uncertainty on the Example of Political Elections

The article presents a system of election recommendation in which both candidate’s and voter’s preferences can be described in an imprecise way. The model of the system is based on IF-set theory which can express hesitation or lack of knowledge. Similarity measures of IF-sets and linguistic quantifiers are used in the decision-making process.

Trust Propagation Based on Group Opinion

Modern technology, especially Internet, allows people to find the resources or knowledge they need by making use of the experiences and opinions of other people. It is easy to collect a vast amount of data, however, a problem of quality and reliability of these data is urgent. Trust networks seem to be the best solution but they need more research and attention. In this paper we join the discussion about trust representation and trust propagation. We follow the idea of modeling trust in gradual and dual form of trust and distrust degrees, using Atanassov’s intuitionistic fuzzy sets (IFSs) theory as the basis. Moreover, we introduce a new trust propagation operator based on group opinion and on relative scalar cardinality of IFSs.

Trust propagation—cardinality-based approach

The article deals with a problem of modelling and propagating trust in trust networks (social networks, recommender systems, integrating systems). We briefly recall some of the solutions and approaches proposed by other authors and we focus on a trust model based on IFS theory (Atanassovs intuitionistic fuzzy set theory). Then we present a draft of a method of calculating trust propagation with a use of a relative cardinality of IFS.

Propagating and aggregating trust with uncertainty measure

Trust networks have been recognized as a valuable component of many modern systems, such as e-commerce or recommender systems, as they provide a way of quality assessment. In addition to adequate modeling of trust in such network, two fundamental issues need to be addressed: the methods of propagation and aggregation of trust. In this paper we present an operator that performs both propagation and aggregation of trust. Trust is modeled on the basis of IFS theory (Atanassovs intuitionistic fuzzy set theory) with particular emphasis on uncertainty, and the operator is based on relative scalar cardinality of IFS. The operator can be used in a very flexible manner for prediction of local and global trust.

Uncertainty-Preserving Trust Prediction in Social Networks

The trust metric has became an increasingly important element of social networks. While collecting, processing and sharing information is becoming easier and easier, the problem of the quality and reliability of that information remains a significant one. The existence of a trust network and methods of predicting trust between users who do not know each other are intended to help in forming opinions about how much to trust information from distant sources. In this chapter we discuss some basic concepts like trust modeling, trust propagation and trust aggregation. We briefly recall recent developments in the area and present a new approach that focuses on the uncertainty aspect of trust value. To this end we utilize a theory of incompletely known fuzzy sets and we introduce a new, uncertainty-preserving trust prediction operator based on group opinion and on relative scalar cardinality of incompletely known fuzzy sets. Motivated by the need for proper uncertainty processing, we have constructed a new method of calculating relative scalar cardinality of incompletely known fuzzy sets that ensures the monotonicity of uncertainty. We outline the problem of uncertainty propagation, and we illustrate by examples that the proposed operator provides most of the desirable properties of trust and uncertainty propagation and aggregation.