Visualization and recovery of the (bio)chemical interesting variables in data analysis with support vector machine...

Area of Science:

• Chemometrics
• Bioinformatics
• Data Science

Background:

• Support Vector Machines (SVMs) are widely used for complex data classification due to their ability to model nonlinear relationships.
• Mapping data to higher-dimensional spaces in SVMs can obscure the contribution of original variables to classification.
• Understanding variable importance is crucial for interpreting results in fields like metabolomics.

Purpose of the Study:

• To introduce an innovative method for retrieving and visualizing the contribution of original variables in Support Vector Machine (SVM) classifications.
• To address the information loss regarding variable contributions inherent in SVM's higher-dimensional mapping.
• To enhance the interpretability of complex data analyses in chemometrics and bioinformatics.

Main Methods:

• Development of a novel method to determine variable contributions within SVM classification models.
• Application of the proposed method to benchmark datasets.
• Validation using a metabolomics dataset to demonstrate practical utility.

Main Results:

• The proposed method successfully retrieves and quantifies the contribution of original variables in SVM classifications.
• Visualization techniques were employed to illustrate variable contributions effectively.
• The method proved effective on both general benchmark datasets and a specific metabolomics dataset.

Conclusions:

• The developed method provides a valuable tool for understanding variable importance in SVM analyses.
• Visualizing variable contributions aids in deciphering complex chemical or biological processes.
• This approach enhances the interpretability and applicability of SVMs in scientific research.