Advanced computational approaches for predicting sunflower yield: Insights from ANN, ANFIS, and GEP in normal and salinity stress environments

Prediction of crop yield is essential for decision-makers to ensure food security and provides valuable information to farmers about factors affecting high yields. This research aimed to predict sunflower grain yield under normal and salinity stress conditions using three modeling techniques: artificial neural networks (ANN), adaptive neuro-fuzzy inference system (ANFIS), and gene expression programming (GEP). A pot experiment was conducted with 96 inbred sunflower lines (generation six) derived from crossing two parent lines, over a single growing season. Ten morphological traits—including hundred-seed weight (HSW), number of leaves, leaf length (LL) and width, petiole length, stem diameter, plant height, head dry weight (HDW), days to flowering, and head diameter—were measured as input variables to predict grain yield. Salinity stress was induced by applying irrigation water with electrical conductivity (EC) levels of 2 dS/m (control) and 8 dS/m (stress condition) using NaCl, applied after the seedlings reached the 8-leaf stage. The GEP model demonstrated the highest precision in predicting sunflower grain yield, with coefficient of determination (R2) values of 0.803 and 0.743, root mean squared error (RMSE) of 4.115 and 4.022, and mean absolute error (MAE) of 3.177 and 2.803 under normal conditions and salinity stress, respectively, during the testing phase. Sensitivity analysis using the GEP model identified LL, head diameter, HSW, and HDW as the most significant parameters influencing grain yield under salinity stress. Therefore, the GEP model provides a promising tool for predicting sunflower grain yield, potentially aiding in yield improvement programs under varying environmental conditions.