Exploratory Projection Pursuit

Abstract A new projection pursuit algorithm for exploring multivariate data is presented that has both statistical and computational advantages over previous methods. A number of practical issues concerning its application are addressed. A connection to multivariate density estimation is established, and its properties are investigated through simulation studies and application to real data. The goal of exploratory projection pursuit is to use the data to find low- (one-, two-, or three-) dimensional projections that provide the most revealing views of the full-dimensional data. With these views the human gift for pattern recognition can be applied to help discover effects that may not have been anticipated in advance. Since linear effects are directly captured by the covariance structure of the variable pairs (which are straightforward to estimate) the emphasis here is on the discovery of nonlinear effects such as clustering or other general nonlinear associations among the variables. Although arbitrary nonlinear effects are impossible to parameterize in full generality, they are easily recognized when presented in a low-dimensional visual representation of the data density. Projection pursuit assigns a numerical index to every projection that is a functional of the projected data density. The intent of this index is to capture the degree of nonlinear structuring present in the projected distribution. The pursuit consists of maximizing this index with respect to the parameters defining the projection. Since it is unlikely that there is only one interesting view of a multivariate data set, this procedure is iterated to find further revealing projections. After each maximizing projection has been found, a transformation is applied to the data that removes the structure present in the solution projection while preserving the multivariate structure that is not captured by it. The projection pursuit algorithm is then applied to these transformed data to find additional views that may yield further insight. This projection pursuit algorithm has potential advantages over other dimensionality reduction methods that are commonly used for data exploration. It focuses directly on the “interestingness” of a projection rather than indirectly through the interpoint distances. This allows it to be unaffected by the scale and (linear) correlational structure of the data, helping it to overcome the “curse of dimensionality” that tends to plague methods based on multidimensional scaling, parametric mapping, cluster analysis, and principal components.