Using auto-associative neural networks to compress and visualize multidimensional data

Abstract

To interpret the information hidden in multidimensional data can be considered as challenging and complicated task. Usually, dimension reduction or data compression is considered as the first step to data analysis and exploration of multidimensional data. Here, the focus is given to study Auto-Associative Neural Networks (AANNs) technique for data compression and visualization. AANNs have the ability to deal with linear and nonlinear correlation among variables. This technique is often referred to as nonlinear Principal Component Analysis (NLPCA) or could also be known as Bottleneck Neural Networks (BNNs) due to their specific structures that consist of combination of compression and decompression networks. The trained AANNs can reduce high dimensional data onto lower dimensional data by compressing them on its bottleneck layer that later can be used for data visualization. In this paper, the technique of AANNs are described, developed using high level computer language and applied on two different multidimensional datasets. The results have shown that AANNs are able to compress multidimensional data into only two non linear principal components at its bottleneck layer and these compressed data can provide visualization of different clusters of data.