Machine Learning in the Context ofLaser‐InducedBreakdown Spectroscopy

This chapter presents the fundamental ideas behind the most common machine-learning (ML) techniques found in the laser-induced breakdown spectroscopy literature. It describes random forests, support vector machines, artificial neural networks, unsupervised learning, and self-organizing maps. The chapter begins, for historical reasons, with one of the first ML algorithms – decision trees. Using the concept of decision trees, it then conceptually introduces several ensemble methods, i.e., methods that combine or aggregate the results of multiple simple models to form a more powerful prediction. Namely, the chapter discusses bootstrap aggregation (bagging), boosting and its more powerful variant, gradient boosting, and lastly, random forests. It emphasizes that while the ensemble methods are described considering decision trees, they are not limited to trees. Thus, the presented ensembling methods can be applied to improve the performance of any other ML model.