GA-Based Feature Selection for Soil Quality Classification Using Soil and Climate Variables

This study investigates genetic algorithm (GA)-based feature selection for soil quality classification in smart agriculture using a synthetic soil-climate dataset. Rather than employing GA as the classifier itself, the method uses GA to optimize the subset of predictors supplied to a class-balanced logistic regression model. Each chromosome encodes the inclusion or exclusion of 10 soil and climate variables, while the fitness function combines F1-score, balanced accuracy, and a sparsity bonus to reward compact solutions under class imbalance. With a population size of 24 and 20 generations, the evolutionary search identified Irrigation_Available and Soil_Nitrogen as the best two-feature subset. On the hold-out test set, the GA-selected model achieved an F1-score of 0.2934, nearly identical to the full-feature baseline (0.2941), while improving balanced accuracy from 0.7950 to 0.8159. These results indicate that GA-based feature selection can reduce model complexity substantially without materially degrading predictive performance. The findings support the use of evolutionary optimization as an interpretable decision-support layer for data-driven agricultural analytics.