Enhancing Fertilization Strategies through Graph Attention Network - Transformer Fusion Model for Smart Farming

India’s GDP is driven by the agriculture sector, which provides numerous individuals with livelihoods. During harvest, climatic and weather conditions highly affect crop production, resulting in losses, and incorrect analysis of these factors can result in lower yields. A well-defined process is needed to develop a model keeping geographical diversity in mind while ensuring accurate, cost-effective fertilization methods. This study introduces a hybrid model by integrating Transformers with Graph Attention Networks (GAT) to estimate fertilizer needs based on the region's unique requirements. GATs identify spatial correlations by modelling farms as graph nodes and edges connected by geographic distance based on closeness. Transformers handle sequential data to reveal temporal patterns. The hybrid model successfully combines spatial-temporal data, identifying complex relationships to make specific fertilization recommendations dynamically. It surpasses conventional ML models' accuracy, scalability, and adaptability, delivering consistent outcomes in the analysis of Tamil Nadu and Punjab regions. As India's agriculture is diverse regarding soil types, climates, and agricultural practices, this adaptive method updates recommendations dynamically, improving precision and relevance for farmers.