Abstract - Classifier ensembles form an important approach to improving classification performance. As such, there have been different proposals made in the literature that provide a range of means to construct and aggregate classifier ensembles. However, the resulting systems may contain unreliable members with false or biased judgements in the ensemble. The removal of unreliable members is necessary to optimise the overall performance of such systems. Smaller ensembles involving reduced ensemble members also helps relax the requirement of computational memory, thereby strengthening the system's run-time efficiency. To differentiate the potential contributions of different ensemble members while reducing the adverse impact of any unreliable judgement upon the system, a nearest neighbour-based reliability measure is incorporated into the process of classifier ensemble selection. In particular, reliabilities of selected ensemble members are perceived as a stress function, from which argument-dependent weights are heuristically generated for final aggregated decision. Experimental investigations are carried out, demonstrating the efficacy of the proposed approach, where fuzzy classifiers are utilised as base members of the emerging ensemble.
Abstract - In this paper interval-valued fuzzy relations are studied. The problem of selection of alternatives in decision making is examined. In particular, a new definition of transitivity based on the measure of the intensity preference between pairs of alternatives is presented and its apply to solve decision making problem is proposed.
Abstract - This contribution is devoted to a method for forecasting of the trend-cycle of time series based on the application of a combination of the higher degree fuzzy transform and fuzzy natural logic (FNL) techniques. The main idea is to predict future components of the direct fuzzy transform of a time series. The forecast trend-cycle is then computed using the inverse fuzzy transform applied to the predicted components.
Abstract - In this paper the problem of connections between input fuzzy relations R1,..,Rn on a set X and the output fuzzy relation RF = F(R1,...,Rn) on X is studied, where F is a function on the unit interval [0,1] and RF is an aggregated fuzzy relation. Namely, fuzzy relation RF = F(R1,...,Rn) is assumed to have a given property and the properties of fuzzy relations R1,...,Rn are examined. This approach to checking connections between input fuzzy relations and the output fuzzy relation is a new one. In the literature the problem of preservation, by an aggregation function F, diverse types of properties of fuzzy relations R1,...,Rn is examined. The properties, which are examined in this paper, depend on their notions on binary operations B on the unit interval [0,1], i.e. they are generalized versions of known properties of fuzzy relations.
Abstract - Supervised learning is of key interest in data science. Even though there exist many approaches to solving, among others, classification as well as ordinal and standard regression tasks, most of them output models that do not possess useful formal properties, like nondecreasingness in each independent variable, idempotence, symmetry, etc. This makes them difficult to interpret and analyze. For instance, it might be impossible to determine the importances of individual features or to assess the effects of increasing the values of predictors on the behavior of a chosen response variable. Such properties are especially important in software plagiarism detection, where we are faced with the combination of degrees to which how much a code chunk A is similar to (or contained in) B as well as how much B is similar to A. Therefore, in this paper we consider a new method for fitting B-spline tensor product-based aggregation functions to empirical data. An empirical study indicates a highly competitive performance of the resulting models. Additionally, they possess an intuitive interpretation which is highly desirable for end-users.
Abstract - In this paper we thoroughly investigate various OWA-based linkages in hierarchical clustering on numerous benchmark data sets. The inspected setting generalizes the well-known single, complete, and average linkage schemes, among others. The incorporation of weights into the cluster merge procedure creates an opportunity to make use of experts' knowledge about a particular data domain so as to generate partitions of a given data set that better reflect the true underlying cluster structure. Moreover, we introduce a correction for the inequality of cluster size distribution - similar to the one proposed in our recently introduced Genie algorithm - which results in a significant performance boost in terms of clustering quality.