Hackem-LIBS: An Heterogeneous Stacking Ensemble Model for Laser-Induced Breakdown Spectroscopy Elemental Quantitative Analysis

Laser-Induced Breakdown Spectroscopy (LIBS) is a popular technique for elemental quantitative analysis in chemistry community, based on which, various methods are developed to determinate the concentrations of chemical samples. Despite the successful applications of the existing methods, they still struggle to obtain accurate samples analyses, due to their limited prediction capability, the complex compositions of samples and mutual interference of elements. In this paper, we propose a novel heterogeneous stacking ensemble learning model called Heterogeneous stACKing Ensemble Model LIBS (Hackem-LIBS) to achieve LIBS quantitative analysis with higher accuracy. Specifically, we propose a stacking ensemble learning framework consisting two stages. In the first stage, we train different heterogeneous component learners with multiple sub-training sets and pick out the optimal learners. In the second stage, we leverage the enhanced features predicted by the selected learners to train a stronger meta-learner, which is used to make the final prediction. In addition, we combine Genetic Algorithm (GA) with Sequential Forward Selection (SFS) to reduce the redundancy of training features, which ensures more effective learning and higher computation efficiency. Extensive experiments on two public benchmarks are conducted and the results show that our approach achieves better accuracy in determinating the concentrations of elements and is practically applicable to the quantitative analysis of complex chemical samples via the LIBS technique.