Sustainable Phytoprotection: A Smart Monitoring and Recommendation Framework Using Puma Optimization for Potato Pathogen Detection

Amal H. Alharbi¹Faris H Rizk²Khaled Sh Gaber³Marwa M. Eid^4*El-Sayed M. El-kenawy^5*PK Dutta⁶Doaa Sami Khafaga¹

• ¹Department of Computer Sciences, College of Computer and Information Sciences, Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia
• ²Department of Communications and Electronics, Delta Higher Institute of Engineering and Technology, Mansoura, 35111, Egypt, Mansoura, Egypt
• ³Computer Science and Intelligent Systems Research Center, Blacksburg 24060, Virginia, USA, Virginia, Minnesota, United States
• ⁴Faculty of Artificial Intelligence, Delta University for Science and Technology, Mansoura, Egypt, Mansoura, Egypt
• ⁵School of ICT, Faculty of Engineering, Design and Information Communications Technology (EDICT), Bahrain Polytechnic, PO Box 33349, Isa Town, Bahrain, Bahrain, Bahrain
• ⁶School of engineering and technology, Amity University, Kolkata, India, Kolkata, India

Ensuring sustainable and resilient agricultural systems in the face of intensifying crop disease threats requires intelligent, data-driven tools for early detection and intervention. This study proposes a novel hybrid framework for potato disease classification that integrates copula-based dependency modeling with a Restricted Boltzmann Machine (RBM), further enhanced through hyperparameter tuning using the biologically inspired Puma Optimization (PO) algorithm. The system is trained and evaluated on a real-world dataset derived from structured field experiments, 1 Alharbi et al.comprising 52 instances and 42 agronomic, microbial, and ecological variables. By fusing copulabased transformations with PO-driven optimization, the framework effectively models complex nonlinear dependencies among heterogeneous features, enabling high-fidelity probabilistic inference in high-dimensional ecological spaces. The RBM baseline outperformed conventional classifiers such as KNN, Random Forest, XGBoost, and MLP, achieving 94.77% accuracy. With PO-based optimization, performance improved significantly to 98.54% accuracy, with parallel gains in sensitivity, specificity, and F1-score. Statistical analysis using ANOVA and Wilcoxon signed-rank testing confirmed the significance of these improvements (p <0.002). In contrast, convergence analysis demonstrated PO-RBM's computational efficiency relative to PSO, GWO, and GA alternatives. These findings underscore the utility of the proposed framework as a scalable and ecologically grounded decision-support system for integrated pest management (IPM), offering a practical path toward low-impact, adaptive plant health monitoring solutions.