fcvalid: Olasılıklı ve Olabilirlikli Bölümleyici Kümelemede Bulanık Geçerlilik İndeksleri için Bir R Paketi

Abstract

In exploratory data analysis and machine learning, partitioning clustering is a frequently used unsupervised learning technique for finding the meaningful patterns in numeric datasets. Clustering aims to identify and classify the objects or the cases in datasets in practice. The clustering quality or the performance of a clustering algorithm is generally evaluated by using the internal validity indices. In this study, an R package named 'fcvalid' is introduced for validation of fuzzy and possibilistic clustering results. The package implements a broad collection of the internal indices which have been proposed to validate the results of fuzzy clustering algorithms. Additionally, the options to compute the generalized and extended versions of the fuzzy internal indices for validation of the possibilistic clustering are also included in the package.

Two novel outlier detection approaches based on unsupervised possibilistic and fuzzy clustering

“…(10) , we divided the typicality degrees into their row sums and obtained the normalized typicality degrees ( u’ ik ) to use in validation of clustering results. In this study, the R package ‘fcvalid’, Cebeci (2020) has been run for computing the internal validity indices using the UPFC clustering results.…”

“…According to the surveys, various methods and approaches have been proposed for outlier detection ( Patcha & Park, 2007 ; Chandola, Banerjee & Kumar, 2009 ; Gogoi et al, 2011 ; Zhang, 2013 ; Kalinichenko, Shanin & Taraban, 2014 ). All these methods and approaches can be grouped in different taxonomies such as univariate vs multivariate; parametric, semi-parametric vs non-parametric; supervised, semi-supervised vs unsupervised; or more frequently in the classes of distribution-based, depth-based, distance-based, density-based and clustering-based methods ( Ben-Gal, 2005 ; Cebeci, 2020 ; Zhang, Meratnia & Havinga, 2010 ).…”

“…In general, they are based on statistical tests, and their efficiencies highly depend on whether the assumptions of statistical models fit to the real data or not. These methods generally cannot be applied on multidimensional data sets, and they also do not work for data sets in which there is no prior information about their distribution ( Ben-Gal, 2005 ; Cebeci, 2020 ).…”

“…On the other hand, density-based proximity methods treat a data point as an outlier if it is located in a low-density region in a data set. In outlier detection, the overall performance of a proximity-based method can be easily affected by the metrics used with it ( Cebeci, 2020 ). The advantage of these methods over statistical-based methods is that no prior assumptions are made about data distribution.…”

Soft clustering

“…All the methods are implemented in R packages. In particular, function FKM of the R package fclust (Ferraro & Giordani, ) has been used for the F k M, function pcm of the R package ppclust (Cebeci, Yildiz, Kavlak, Cebeci, & Onder, ) for the P k M, function RoughKMeansLW of the R package softclustering (Peters, ) for the R k M and function Mclust of the package mclust (Scrucca, Fop, Murphy, & Raftery, ) for FMG. The dataset contains 200 objects, characterized by two clusters.…”