Deep autoencoder-driven feature learning and meta-heuristic optimized machine learning modelling for crop water stress identification

Accurate identification of water stress in crops is essential for maintaining sustainable, high-quality crop production and minimizing the risk of severe crop losses. The erroneous and subjective nature of human expert-based crop stress identification can be effectively replaced by machine learning...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Evolving systems Jg. 16; H. 3; S. 108
Hauptverfasser: Subeesh, A., Chauhan, Naveen, Chandel, Narendra Singh, Rajwade, Yogesh
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Berlin/Heidelberg Springer Berlin Heidelberg 01.09.2025
Springer Nature B.V
Schlagworte:
Accuracy
Agricultural production
Applications programs
Artificial Intelligence
Climate change
Complex Systems
Complexity
Crop production
Data analysis
Data processing
Decision support systems
Deep learning
Engineering
Heuristic
Heuristic methods
Identification
Machine learning
Mobile computing
Optimization techniques
Original Paper
Performance enhancement
Physiology
Prediction models
Voting
Water
Deep learning
Precision agriculture
Ensemble model
Autoencoders
Abiotic stress
ISSN:1868-6478, 1868-6486
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Accurate identification of water stress in crops is essential for maintaining sustainable, high-quality crop production and minimizing the risk of severe crop losses. The erroneous and subjective nature of human expert-based crop stress identification can be effectively replaced by machine learning (ML) and deep learning models (DL) due to their remarkable ability to perform complex data analysis. The objective of the present study is to develop a framework for the accurate identification of crop water stress using a hybrid DL-ML approach. The study proposes a novel method utilizing the feature representation capabilities of a sparse autoencoder (SAE) to extract latent space features and optimize the ML models (XGB, RF, LGB) using the Bat Algorithm to create a strong predictive model through a soft voting ensemble (SAE-BAT-ENS). The performance of optimized models viz., SAE-BAT-XGB, SAE-BAT-RF, and SAE-BAT-LGB were compared with the state-of-the-art (SOTA) models. Although the SAE-BAT-XGB model showed strong predictive power with an accuracy 97.08%, the soft voting ensemble model, SAE-BAT-ENS, enhanced the performance, by taking the prediction probabilities into account, thus balancing out the weaknesses of individual models. Additionally, the proposed approach was compared with the standard majority voting ensemble (MAJ_VOTE_ENS), and the SAE-BAT-ENS achieved superior results. The proposed approach outperformed the SOTA ML models with an accuracy of 97.81%, precision 98.50%, recall 97.05%, and F1-score 97.77%. The approach employed in the study has shown an increment of 3.94% compared to RF, 3.01% compared to model XGB, and 4.71% compared to LGB model. The study finds applications in developing precision crop water stress management, decision support systems, and mobile applications for automating field crop management.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1868-6478
1868-6486
DOI:10.1007/s12530-025-09729-2