Climate change based yield production using hybrid deep learning with optimal feature selection

Abstract

Accurate agricultural yield prediction under changing climate conditions has become the challenge task due to the increasing variability in environmental factors like temperature fluctuations, irregular rainfall patterns and soil degradation. Traditional statistical and machine learning approaches have issues in capturing the complex nonlinear relationships and temporal dependencies among the climate-driven agricultural data. To address the limitations, this research proposed the novel Climate Change based Yield Production using Hybrid Deep Learning (CY-HDLNet) model with data normalization, optimal feature selection and deep learning based yield prediction. Initially, log transformation is employed to standardize skewed climate data distributions and feature selection is employed using the proposed Chaotic Basketball Team Optimization (CBtO) algorithm that integrates Chebyshev chaotic mapping to enhance global search and avoid local optima. The acquired optimal best feature is then processed using the proposed Recurrent Dual Attention Transformer (RDAT) model with dual attention mechanism to capture both feature-level and temporal dependencies and recurrent unit for improving the sequential learning capability. The experimental results demonstrate that the proposed model significantly enhances prediction accuracy and convergence speed for climate-resilient agricultural planning and decision-making.