REPPlab: An R package for detecting clusters and outliers using exploratory projection pursuit

Abstract

The R-package REPPlab is designed to explore multivariate data sets using onedimensional unsupervised projection pursuit. It is useful in practice as a preprocessing step to find clusters or as an outlier detection tool for multivariate numerical data. Except from the package tourr that implements smooth sequences of projection matrices and rggobi that provides an interface to a dynamic graphics package called GGobi, there is no implementation of exploratory projection pursuit tools available in R especially in the context of outlier detection. REPPlab is an R interface for the Java program EPP-lab that implements four projection indices and three biologically inspired optimization algorithms. The implemented indices are either adapted to cluster or to outlier detection and the optimization algorithms have at most one parameter to tune. Following the original software EPP-lab, the exploration strategy in REPPlab is divided into two steps. Many potentially interesting projections are calculated at the first step and examined at the second step. For this second step, different tools for plotting and combining the results are proposed with specific tools for outlier detection. Compared to EPP-lab, some of these tools are new and their performance is illustrated through some simulations and using some real data sets in a clustering context. The functionalities of the package are also illustrated for outlier detection on a new data set that is provided with the package.

Large-sample properties of unsupervised estimation of the linear discriminant using projection pursuit

“…Similarly one could extend our considerations here to many other PP indices as well, which often are modifications of skewness or kurtosis (see e.g. [25]) or otherwise motivated to be useful in clustering or structure detection, see, for example, [14,17] and references therein for alternative indices. These indices are however often computationally expensive and therefore much less popular than skewness and kurtosis.…”

ICS for multivariate outlier detection with application to quality control

“…The quality standards for this process are respected for each variable, and the objective is to detect some potential multivariate faulty units representing less than 2% of the 520 observations. In Fischer et al (2016), two observations (414 and 512) are detected as the most severe outliers.…”

“…In Table 4, the results for the MCD are not reported. As mentioned in Fischer et al (2016), computing the MCD (at least with a breakdown point equal or larger than 25%)…”

Non-Gaussian Component Analysis: Testing the Dimension of the Signal Subspace

“…. , d p correspond to kurtosis measures of latent variables z yielding d i = 1 if and only if E(z 4 i ) = 3. Thus, in ICA, the FOBI functional is well-defined (up to signs) if all independent components have distinct kurtoses and in that case z corresponds to the original independent components up to signs and order.…”