Seed yield optimization in soybean: integrating biochemical and environmental factors using TabPFN for precision agriculture

Abstract

Precision agriculture seeks yields and optimizes the value of resources consumed by data-driven decisions. Soybean is an important crop worldwide, significant for food and industrial use. Traditional methods fail to accurately predicting seed yield because plant growth is complex, determined by the genotype. This study is to develop a deep learning architecture to achieve optimum soybean growth and seed yield by using treatment by salicylic acid, water stress, and genotype through key biochemical traits for advancing precision agriculture. Initially, Soybean Agricultural Dataset is collected and cleaned by decoding experimental parameters and converting to one-hot encoding to form interpretable treatment groups and handle missing values by mean imputation. Moreover, Biochemical trait engineering composite indices as Drought Tolerance Index and Yield Efficiency Ratio be incorporated to extract the biochemical features. The extracted biochemical traits support plant nutrition by improving the understanding of plant physiology, plant growth, and nutrient requirements in agronomic crops, enabling precision agriculture for enhanced soybean crop yield. The Analysis of Variance (ANOVA) F-test was used for feature selection to detect the most important yield determinants. Afterward, a Tabular Prior-Data Fitted Network model was constructed for the projections of seed yield and confidence interval calculations. The proposed model resulted in a mean square error of 0.07564 and an R2 score of 98.54%, hence it was recognized as a marvelous model regarding accuracy. This will bring about sustainable agricultural practices and better choices in farms.