Precision Agriculture Using a Two-Tier ML Model: Integrating aKNCN Soil Classification with ELM-mBOA Yield Prediction

Abstract

Accurate crop yield prediction is critical for sustainable agricultural planning and resource optimization, especially amidincreasing food demand and climate variability. This study proposes a novel two-tiered machine learning (ML) frameworkthat integrates IoT-based soil data with advanced classification and regression models to enhance prediction accuracy. Inthe first tier, an Adaptive k-Nearest Centroid Neighbour (aKNCN) classifier evaluates soil quality based on key nutrientmetrics. The second tier utilizes an Extreme Learning Machine (ELM) optimized via the modified Butterfly OptimizationAlgorithm (mBOA) to forecast crop yields, incorporating both soil quality and agro-environmental factors. The systemis trained and validated on a publicly available Indian crop production dataset containing 10,000 samples across majorcrops (wheat, maize, rice), with features including soil moisture, temperature, and rainfall. Feature selection is performedusing Correlation-Based Feature Selection (CBFA) and Variance Inflation Factor (VIF) methods to reduce noise and mul-ticollinearity. Experimental results demonstrate that the proposed aKNCN-ELM-mBOA model significantly outperformstraditional ML models—such as Support Vector Machine (SVM), Artificial Neural Network (ANN), Gradient Boosting(GB), and Random Forest (RF)—in terms of error metrics including Mean Absolute Error (MAE), Root Mean SquaredError (RMSE), Mean Absolute Percentage Error (MAPE), and R². The model achieves a notably low RMSE of 0.301and MAPE of 3.932, alongside a high R² score of 0.817, indicating strong generalization. This approach underscores thepotential of hybrid ML systems, enriched by IoT-driven data and robust optimization, to drive precision agriculture andinformed decision-making. Future work may involve time series forecasting and scaling the model with real-time sensordata for broader deployment.