Genetic Algorithm-Optimized Machine Learning Models for Soil Health Assessment in Dryland Wheat Cultivation

The development of high-quality, edible wheat depends on accurate soil health assessments conducted during dryland wheat farming. Limitations in their practical application are often caused by the absence of adequate calibration and validation in traditional methods used to evaluate soil quality. There are many physical, chemical, and biological factors that affect soil health; therefore, good indicators need to be quantifiable and adaptable to management strategies. Soil is inherently dynamic, hence this study set out to create a more precise and flexible method of evaluation. They tested four preprocessing approaches with 96 spectral indices: CR, FD, SNV, and 3SV. To ensure the data was accurate, Missing Value Imputation was used. A hybrid model was created to construct soil health surrogates. It used SVM for feature selection and GA for model optimisation. With reasonable values for MAE (0.026), RRMSE (10.467), R² (0.543), and r (0.687), the SVM-GA model showed promise. Based on these results, it seems like the SVM-GA method can be used to evaluate the soil's health in dryland wheat systems. This would be a great step forward for agricultural soft computing and sustainable soil management.