A novel method for tracing gasoline using GC-IRMS and Relief-Stacking fusion model
Zhaowei Jie,Xiaohan Zhu,Hanyu Zhang,Hanyang Zheng,Can Hu,Zhanfang Liu,Hongling Guo,Jun Zhu,Jianghai Lu,Hongcheng Mei
DOI: https://doi.org/10.1016/j.microc.2024.112081
IF: 5.304
2024-11-13
Microchemical Journal
Abstract:Due to China's vast geographical expanse and the complexity of its natural environment, traditional methods are not suitable for tracing the origins of gasoline samples from different regions of China. The aim of this study is to trace the origins of gasoline samples through the analysis of carbon stable isotopic characteristics from 60 gasoline samples collected across four major regions in China. In this study, a novel method that combines gas chromatography-isotope ratio mass spectrometry (GC-IRMS) with Relief-Stacking was proposed for the first time, significantly enhancing the precision of the provenance results. Firstly, eight characteristic components were identified by Relief feature selection model for the first time and applied to the assessment of carbon stable isotope ratios across different regions. The discovery of these components effectively reduces data redundancy, thereby enhancing the accuracy of provenance analysis. Subsequently, five effective source tracing models (XGBoost, Random Forest, CatBoost, LightGBM, and KNN) were selected based on the CEI values. To further enhance the learning efficiency of the sourcing models, this study integrated the five selected source tracing models using three different methods(voting, stacking, and blending) and validated the performance of the optimal sourcing model. The results indicated that the Stacking fusion model achieves a CEI value of 0.923, significantly outperforming other fusion models. This advantage is attributed to the Stacking fusion model combining the s trengths of Boosting and Bagging ensemble methods and leveraging the advantages of individual classifiers within the model. The method proposed in this study can effectively address the limitations of previous tracing methods and offer valuable insights for the development of platforms related to stable isotope source tracing. In future research, through appropriate parameter adjustments and feature adaptations, the Relief-Stacking model can likewise be applied to source tracing studies in other countries characterized by extensive geographic ranges and variable environmental conditions.
chemistry, analytical