Abstract :

The Increasing Need Of Sustainable Food Production Requires The Use Of Smart Decisionsupport Systems That Are Able To Successfully Categorize The Types Of Crops And Forecast The Level Of Yield Based On The Data Of Heterogeneous Agricultural Sources. The Given Paper Introduces A Novel Hybrid Machine Learning Model, That Is, A Gradient Boosted Trees (GBM) And Deep Neural Network (DNN) That Are Used Simultaneously To Perform Multi-crop Classification And Yield Prediction Based On Data. The Proposed Architecture Uses Two Stage Pipeline Of Learning: The First Stage Of Learning A Bagged Ensemble Of Decision Trees (GBM) Is Applied To A Fused Feature Space Of Normalized Continuous Agronomic Parameters, Such As Soil Nutrients, Water Availability, Temperature And Humidity, And The Second Stage Is One-hot Encoded Categorical Variables, Such As Crop Type And Growing Season. The Posterior Class Probabilities That Are Created By The GBM Are Then Combined With The Original Feature Vector To Create An Enriched Hybrid Input Representation. A Regularized DNN With Fully Connected Layers, ReLU Activations, Dropout Regularization And Softmax Classification Is Trained In The Second Step To Learn Higher-order Feature Interactions Using This Augmented Representation To Allow Finer Discrimination Between Classes Of Crop Yield (Low, Medium, High). A Parallel GBM Regression Model With LSBoost Is Further Estimated To Give Continuous Numeric Yield Predictions Which Are Compared With RMSE, MAE, R 2 And MAPE. The Stratified Holdout Validation Protocol Is 7030 And This Guarantees The Objective Evaluation Of Generalization. Extensive Experimental Analysis Of A Multi-crop Agricultural Dataset Proves That The Proposed Hybrid Model Has Better Classification Performance Compared To Individual Models, And The Precision, Recall, And F1-scores Are Balanced Across All Yield Classes. The Importance Of Features Analysis Also Helps To Identify The Most Significant Agronomic Drivers, Which Can Be Used In Practice As A Solution In Managing Crops And Resources On The Farm Level. A Graphical User Interface Based Inference System Is Also Designed To Enable The Real-time Yield Prediction To The End Users. It Is Indicated In The Proposed Framework That Ensemble Learning Paired With Deep Neural Architectures Has The Potential To Transform Intelligent Agricultural Systems To Achieve Scalability, Interpretability, And High Performance.

Published:

24-4-1-2026

Issue:

Vol. 26 No. 4-1 (2026)

Page Nos:

906-920

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite