Understanding Australian illicit drug markets: unsupervised learning with independent component analysis

The paper proposes a decision rule for separating unlabeled data into two contrasting populations. The driving problem is to understand the Australian illicit drug market based on 17 key indicator data series. A classification based on the most non-Gaussian data direction is proposed. This direction is determined using independent component analysis. It is shown that the resulting grouping leads to interpretable and meaningful measures for describing the drug market.