Metaheuristic-Enhanced Dual Attention Networks for Predictive Analysis of Carbon Sequestration Dynamics
Year: 2026
Type: conference-paper
Abstract: To create a framework that uses a variety of environmental variables to estimate Carbon Sequestration (CS) with a high accuracy as well as computing efficiency. A data-processing model, involves null-value and outlier detection, exploratory analysis to investigate feature behaviour and relationships. Feature engineering guarantees clean and organised input for model training through label encoding and normalisation. The proposed Dual Attention Vision Transformer (DA-ViT), which uses dual-attention mechanisms to capture compound spatial as well as temporal patterns found in carbon sequestration datasets. To increase model performance; hyper parameters are optimized using Green Anaconda Optimization (GAO) method, offering effectual global search capability and reliable convergence. For proposed DA-ViT with GAO the error metrics RMSE of 0.0353, MAE of 0.0217, MSE of 0.0012 and $\mathbf{R}^{\mathbf{2}}$ metrics of $98 \%$, illuminating enhanced precision and robustness over older techniques, proven by python software. The proposed DA-ViT with GAO provides a scalable and dependable method to facilitate advanced CS. The strength of the paper is it improves model performance by optimizing hyper parameters through efficient global search and stable convergence behaviour, subsequent in improved robustness and reduced prediction error.
Institution: Chaitanya
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7203869079