Related Experiment Video
Updated: Sep 11, 2025
20:24
Characterization of Complex Systems Using the Design of Experiments Approach: Transient Protein Expression in Tobacco as a Case Study
Published on: January 31, 2014
16.6K
Construction and interpretation of tobacco leaf position discrimination model based on interpretable machine learning
Ranran Kou1, Cong Wang1, Jinxia Liu2
1Key Laboratory of Tobacco Chemistry, Zhengzhou Tobacco Research Institute of China National Tobacco Corporation (CNTC), Zhengzhou, China.
Frontiers in Plant Science
|August 11, 2025
Summary
This study introduces a new machine learning approach for identifying tobacco leaf position using chemical components. The method accurately discriminates leaf positions and reveals the influence of specific chemical compounds.
Area of Science:
- Agricultural Science
- Analytical Chemistry
- Machine Learning
Background:
- Tobacco leaf quality is linked to chemical composition, with leaf position being a key indicator.
- Traditional methods for analyzing chemical composition are time-consuming and labor-intensive.
- Existing near-infrared (NIR) spectroscopy models for leaf position discrimination lack interpretability.
Purpose of the Study:
- To develop an interpretable machine learning model for tobacco leaf position discrimination.
- To identify key chemical components influencing leaf position.
- To integrate rapid chemical analysis with advanced interpretability techniques.
Main Methods:
- Utilized near-infrared rapid analysis to obtain 70 tobacco leaf chemical components.
- Built discrimination models using Support Vector Machine (SVM), Back Propagation Neural Network (BPNN), and Random Forest (RF), optimized with Particle Swarm Optimization (PSO).
- Applied SHapley Additive exPlanations (SHAP) for model interpretability and feature importance analysis.
Main Results:
- The SVM-hybrid kernel model achieved high accuracy: 98.17% on the training set and 96.33% on the test set.
- SHAP analysis successfully ranked feature importance and elucidated the contribution of individual chemical components.
- The study demonstrated the effectiveness of integrating machine learning with interpretability for crop analysis.
Conclusions:
- The proposed method offers a robust and interpretable solution for tobacco leaf position traceability.
- This approach facilitates the identification of critical chemical markers associated with leaf position.
- The methodology holds potential for application in analyzing chemical features across various agricultural crops.
