Related Experiment Video
Updated: Nov 26, 2025
Dependence of Laser-induced Breakdown Spectroscopy Results on Pulse Energies and Timing Parameters Using Soil Simulants
Published on: September 23, 2013
A novel hybrid filter/wrapper method for feature selection in archaeological ceramics classification by laser-induced
Fangqi Ruan1, Lin Hou, Tianlong Zhang
1Key Laboratory of Synthetic and Natural Functional Molecular Chemistry of Ministry of Education, College of Chemistry & Material Science, Northwest University, Xi'an, China. tlzhang@nwu.edu.cn huali@nwu.edu.cn.
A new hybrid feature selection method, MI-DBS, improves qualitative analysis in Laser-Induced Breakdown Spectroscopy (LIBS) for cultural heritage. This approach enhances predictive performance and reduces computational time for complex datasets.
Area of Science:
- Analytical Chemistry
- Spectroscopy
- Cultural Heritage Science
Background:
- Laser-Induced Breakdown Spectroscopy (LIBS) is a powerful analytical technique for cultural heritage.
- High-dimensional LIBS data requires pre-processing for effective multivariate classification.
- Feature selection (FS) is crucial for optimizing LIBS data by removing irrelevant or redundant information.
Purpose of the Study:
- To propose a novel hybrid filter/wrapper feature selection method (MI-DBS) for enhancing qualitative analysis in LIBS.
- To improve the predictive capacity and comprehensibility of multivariate classification models based on LIBS data.
- To reduce computational time and identify optimal feature subsets from complex spectral data.
Main Methods:
- A hybrid feature selection method combining Mutual Information (MI) and Bi-Directional Selection (DBS) algorithms was developed.
- Wavelet Transform Denoising (WTD) was employed to reduce noise in LIBS spectra.
- The proposed method was validated using 35 archaeological ceramic samples and a Random Forest (RF) classifier.
Main Results:
- The MI-DBS algorithm effectively reduced redundant and uncorrelated features from LIBS spectra.
- The hybrid method demonstrated superior predictive performance compared to other feature selection techniques.
- The RF classifier achieved high sensitivity (0.9722), specificity (0.9956), and accuracy (0.9850) using MI-DBS selected features.
Conclusions:
- The MI-DBS algorithm is an effective feature selection strategy for LIBS-based qualitative analysis in cultural heritage.
- This method enhances model performance, reduces feature dimensionality, and decreases computational time.
- MI-DBS offers a valuable alternative for feature selection in complex multivariate classification tasks using LIBS data.
