Quantum-Assisted Hybrid Optimization for Redundancy-Aware Feature Selection to Support Crop Yield Optimization

Abstract

Soil test results, weather observations, remote sensing indices, irrigation records, and field management logs are examples of high-dimensional agricultural data that are increasingly needed for crop production optimization. In addition to increasing training costs and decreasing model stability, redundant or weakly correlated variables may also have less agricultural significance. This paper proposes a quantum-assisted feature selection algorithm for crop yield optimization. Candidate agricultural variables are encoded in quantum probability, feature subsets are assessed using a yield-oriented relevance and redundancy objective, and the evolution of subsets is guided by quantum rotation updating. Agronomic consistency, redundancy suppression, feature sparsity, and yield prediction error are all addressed concurrently by a combined fitness function. The trials included a variety of crop data sources, and the suggested approach outperformed the conventional wrapper selection in terms of prediction accuracy, feature dimensionality reduction by 42.5%, and training time reduction by 27.3%. Based on the findings, precision agriculture and data-driven crop management can be effectively and comprehensibly optimized through the use of quantum-assisted feature selection.