Comparative Evaluation of Allometric, Machine Learning, and Ensemble Approaches for Modeling Dynamic Structure-Fresh...

Area of Science:

• Agricultural Science
• Plant Physiology
• Computational Biology

Background:

• Accurate fresh weight (FW) estimation is vital for greenhouse crop management, aiding growth monitoring and yield prediction.
• Dynamic biomass partitioning in plants presents a significant challenge for precise FW estimation.
• Non-destructive methods are preferred for FW assessment in commercial greenhouse settings.

Purpose of the Study:

• To systematically evaluate and compare different modeling strategies for fresh weight (FW) estimation in greenhouse-grown sweet peppers.
• To identify the most suitable modeling approach for FW prediction under dynamic biomass partitioning conditions.
• To assess the impact of data augmentation techniques on FW estimation accuracy.

Main Methods:

• Established baseline allometric models using non-destructive morphological measurements.
• Developed and compared various machine learning (ML) models, including ensemble frameworks.
• Applied numerical data augmentation using Gaussian noise and a variational autoencoder to address data limitations.
• Analyzed feature contributions to understand predictors for shoot and fruit FW.

Main Results:

• The best allometric model achieved R2 values of 0.80 for shoot FW and 0.54 for fruit FW.
• All ML models surpassed allometric models in predictive accuracy.
• The ensemble ML model demonstrated superior performance, achieving R2 of 0.96 for shoot FW and 0.89 for fruit FW.
• Data augmentation enhanced ML model performance, especially for fruit FW prediction.

Conclusions:

• Ensemble-based machine learning, coupled with data augmentation, offers a robust framework for non-destructive FW estimation in controlled environments like greenhouses.
• Temporal progression is a key factor for fruit FW prediction, while structural traits are dominant for shoot FW estimation.
• This approach supports advanced applications in smart farming and precision agriculture for sweet pepper cultivation.