A feature-centric decision-making framework for diagnosing and enhancing system efficiency in intelligent multi-agent...

Area of Science:

• Manufacturing Systems Engineering
• Artificial Intelligence
• Data Science

Background:

• Industry 4.0 and 5.0 demand advanced analytics for intelligent multi-agent manufacturing.
• Existing predictive models often lack interpretability, resilience, and high performance.
• A feature-centric approach is needed to integrate diverse indicators for system efficiency forecasting.

Purpose of the Study:

• To develop a feature-centric hybrid predictive framework for forecasting system efficiency in intelligent multi-agent manufacturing.
• To enhance the interpretability, resilience, and performance of predictive analytics in Industry 4.0/5.0.
• To integrate operational, learning-based, and cyber-physical indicators for a comprehensive system view.

Main Methods:

• A structured pipeline involving recursive feature elimination for feature selection.
• ANOVA-based sensitivity assessment for statistical variance attribution.
• SHAP for global explainability and integration with tree-based models (decision trees, random forests, CatBoost, extra trees) and meta-heuristic optimizers (prairie dog optimization, electric eel foraging optimization).

Main Results:

• Hybrid models, particularly prairie dog optimization-enhanced random forest and extra trees, significantly improved accuracy, stability, and error reduction.
• Sensitivity analyses identified production efficiency, machine usage, Q-value, and security event as key predictors.
• The feature-driven modeling and bio-inspired optimization approach proved viable for robust and interpretable smart manufacturing applications.

Conclusions:

• The proposed framework offers a novel, explainable, and deployable predictive intelligence paradigm for modern multi-agent industrial systems.
• Feature-centric modeling combined with biologically inspired optimization delivers practical, robust, and interpretable outcomes.
• The research validates the effectiveness of integrating diverse indicators and advanced optimization techniques for industrial performance dynamics.