HCA-DBN: a hill climbing optimized Deep Belief Network for crop yield classification based on kernel weight...

Area of Science:

• Agricultural Science
• Machine Learning
• Data Science

Background:

• Accurate maize yield potential classification is crucial for food security and agricultural planning.
• Environmental variability and socio-economic factors necessitate robust yield prediction methods.
• Maize kernel weight classification (low <25g, high ≥25g) is explored.

Purpose of the Study:

• To develop and evaluate a novel machine learning model for binary classification of maize kernel weight.
• To benchmark the proposed model against standard classification algorithms.
• To establish a methodological framework for field-based maize yield classification.

Main Methods:

• A Hybrid Cascade - Deep Belief Network (HCA-DBN) was proposed, integrating Deep Belief Networks (DBN) for feature extraction and Hill Climbing Algorithm (HCA) for hyperparameter tuning.
• Plant and ear traits from 160 organic maize samples in Tamil Nadu were utilized.
• Performance was compared against Logistic Regression, Random Forest, XGBoost, Decision Tree, MLP, and SVC.

Main Results:

• The HCA-DBN model achieved a peak classification accuracy of 94%.
• The proposed model demonstrated superior performance compared to conventional baseline classifiers.
• Statistical stability was confirmed through bootstrapping and stratified 10-fold cross-validation.

Conclusions:

• The HCA-DBN presents a promising approach for accurate maize yield classification, even with limited data.
• This study provides a proof-of-concept and a scalable framework for future research on larger datasets.
• The findings contribute a methodological benchmark for field-based maize yield prediction.