Study of cluster analysis used in explosives classification with laser-induced breakdown spectroscopy

Supervised learning methods (such as partial least squares regression-discriminant analysis, SIMCA, etc) are widely used in explosives recognition. The correct classification rate may be lowered if a sample or substrate is not included in the training dataset. Unsupervised learning methods (such as hierarchical clustering analysis, K-means, etc) have the potential to solve this problem. In this paper we analyzed results of using as input variables the intensities of seven lines and then five intensity ratios of the seven lines. It was demonstrated that unsupervised learning methods had the ability to achieve a better classification result.