Hybrid enzyme action optimizer CatBoost ensemble for drawbar load prediction with graphical user interface and comparative evaluation against nature inspired metaheuristic algorithms

Abstract Cognitive intelligence of standalone machine learning models is constrained by target leakage, weak data mining, inadequate hyperparameter tuning, and premature convergence to local optima. This study integrated Base-CatBoost Ensemble learner with 25 recent nature-inspired metaheuristic algorithms and evolved new hybrid-CatBoost models for predicting drawbar load (H Db ). These included, Aurora Intelligent Metaheuristic-AIM algorithm, Artificial Lemming Algorithm-ALA, Bobcat Optimization Algorithm-BOA, Bacteria Phototaxis Optimizer-BPO, Chicken Swarm Optimization-CSO, Enzyme Action Optimizer-EAO, Grizzly Bear Fat Increase Optimizer-GBFIO, Gazelle Optimization Algorithm-GOA, Grey Wolf Optimizer-GWO, Harris hawks optimization-HHO, Hippopotamus Optimization Algorithm-HO, Jellyfish Search Optimizer-JSO, Mantis Shrimp Optimization Algorithm-MShOA, Osprey Optimization Algorithm-OOA, Parrot Optimizer-PO, Pufferfish Optimization Algorithm-POA, Reindeer Cyclone Optimization Algorithm-RCOA, Secretary bird optimization algorithm-SBOA, Swift Flight Optimizer-SFO, Snake Optimizer Algorithm-SO, Walrus Optimization-WAOA, Warthog Optimization Algorithm-WartOA, Wolf-Bird Optimizer-WBO, Whale Optimization Algorithm-WOA and White Shark Optimizer-WSO. A scikit-learn library for the Base-CatBoost learner was created, and critical hyperparameters of hybrid ensemble models were heuristically optimized, targeting the experimental H Db obtained from 3360 values in the 14 soil-machine database. Total multi-metric accuracy (R, R 2 , R 2 adjusted, EVS, RAE, RRMSE, RMSRE, NRMSE, SI, PI, WMAPE, VAF, PA, CV) and noise robustness (MSE, RMSE, MAE, MedAE, Huber loss-Lδ, Log-Cosh loss-LCL) score ranking (980), regression error characteristic curve-REC (AOC = 0.236), Taylor reference clustering (R-σ-cRMSE), and convergence stability (perturbation score, 0.005, generalization gap, 4.545e-06) confirmed hybrid EAO-CatBoot as the superior model. SHAP explainable hybrid AI-XHAI revealed that tractive thrust most significantly dominated H Db prediction dispersion and convergence decision path attribution in EAO-CatBoost. Python-based GUI for hybrid EAO-CatBoost is presented for user-friendly HDb prediction in engineering practice.