A Robust Support Vector Machine Model for Monitoring Methane Emission Levels in Paddy Ecosystems using Electronic Nose Technology

Abstract

Paddy farming is a significant contributor to global greenhouse gas emissions, particularly methane (CH₄) and carbon dioxide (CO₂). While accurate monitoring is vital for carbon management, traditional analytical methods are often cost-prohibitive and technically complex. This study develops a robust Support Vector Machine (SVM) model integrated with Electronic Nose (E-Nose) technology to predict carbon emission levels by analyzing multi-dimensional environmental and gas parameters. A comprehensive dataset was collected across three field locations (water inlet, mid-field, and water outlet), comprising CH₄ and CO₂ concentrations, ambient temperature, humidity, and rice growth stages. The data underwent a rigorous pre-processing pipeline, including median imputation for missing values and outlier removal using the Interquartile Range (IQR) method, which refined the dataset from 1,499 to 1,486 instances. Feature selection via the SelectKBest method identified CH₄ and CO₂ concentrations, growth stages, and gas fluxes as the most significant predictors. The SVM model utilizing a Radial Basis Function (RBF) kernel was evaluated using two configurations: "All Features" and "Selected Features". The "Selected Features" model demonstrated superior predictive performance on the testing set, achieving a coefficient of determination R² of 0.9975, a Mean Absolute Error (MAE) of 0.0067, and a Root Mean Square Error (RMSE) of 0.0085. In comparison, the "All Features" model yielded a slightly lower R² of 0.9918, with a higher MAE of 0.0122 and RMSE of 0.0153. This indicates that isolating key environmental drivers not only reduces computational redundancy but also enhances the model's accuracy and generalizability. The results demonstrate that an optimized SVM model, supported by strategic feature selection, provides a highly reliable and cost-effective tool for monitoring carbon footprints in paddy ecosystems. This framework facilitates the transition toward data-driven sustainable farming and precision emission management.