AGRICULTURE DECISION SYSTEM ON NEW MACHINE LEARNING METHODS FOR YIELD PREDICTION

Abstract

ABSTRACT Global food security faces unprecedented challenges from climate change, population growth, and resource limitations, necessitating accurate and scalable crop yield prediction systems. This study presents a comprehensive comparative analysis of contemporary machine learning and deep learning methodologies for agricultural yield prediction, synthesizing findings from 23 peer-reviewed studies published between 2023-2025. Our analysis evaluates model performance across diverse crops, geographical contexts, and data modalities including meteorological parameters, satellite-derived vegetation indices, UAV-based multispectral imagery, and soil properties. Results demonstrate that ensemble methods and hybrid architectures consistently outperform single-algorithm approaches, with Random Forest achieving R² values of 0.875 for Irish potatoes and 0.817 for maize, Support Vector Regression attaining R²=0.95 for wheat yield prediction using UAV multispectral data, and hybrid CatBoost models demonstrating superior performance in capturing complex soil-climate interactions. Deep learning architectures, particularly CNN-SVM hybrids, achieved 97.54% accuracy in tomato grading applications. The findings reveal that no universal model excels across all crops and contexts; rather, model selection must be optimized for specific crop characteristics, data availability, and geographical conditions. We propose a multi-layered agriculture decision system architecture integrating real-time environmental data, remote sensing inputs, and automated machine learning pipelines for scalable yield prediction. Key challenges identified include computational demands, limited model interpretability, and applicability constraints in data-scarce regions. This research contributes empirical benchmarks for model selection and provides a framework for developing context-aware agricultural decision support systems. Keywords: Crop yield prediction; machine learning; deep learning; ensemble methods; remote sensing; decision support system; precision agriculture.