Data-driven PSO-CatBoost machine learning model to predict the compressive strength of CFRP- confined circular concrete specimens

• Published in: Thin-walled structures Vol. 198; p. 111763
• Main Authors: Khodadadi, Nima, Roghani, Hossein, De Caso, Francisco, El-kenawy, El-Sayed M., Yesha, Yelena, Nanni, Antonio
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-05-2024
• Subjects: Carbon Fiber-Reinforced Polymer; Compressive strength; Confined-concrete; Machine learning; PSO-CatBoost; PSO-CatBoost; Compressive strength; Carbon Fiber-Reinforced Polymer; Confined-concrete; Machine learning
• ISSN: 0263-8231; 1879-3223
• DOI: 10.1016/j.tws.2024.111763

•Gathering a comprehensive dataset for modeling the compressive strength of CFRP- Confined Circular Concrete.•Predicting the compressive strength of CFRP- CC specimens using data-driven methods.•Investigation the effects of various factors on the compressive strength.•Hybridization of Particle Swarm Optimization (PSO) and the CatBoost algorithm.
This work articulates the development of a sophisticated machine-learning model for the prediction of compressive strength in Carbon Fiber-Reinforced Polymer Confined-Concrete (CFRP-CC) specimens. Despite extensive empirical studies conducted over the last three decades, prevailing predictive models predominantly rooted in linear or nonlinear regression analyses are constrained by their dependency on limited data scopes. Addressing this deficiency, our research delineates the formulation of an innovative Particle Swarm Optimization- Categorical Boosting (PSO-CatBoost) algorithm, underpinned by an expansive database encompassing 916 experimental outcomes from 105 scholarly articles, spanning the period from 1991 to mid-2023. This innovative approach effectively combines the strengths of Particle Swarm Optimization and the CatBoost algorithm. It carefully evaluates various vital factors that affect the compressive strength of CFRP-CC. The uniqueness of our approach is further accentuated through the application of SHapley Additive exPlanations (SHAP) and Permutation Feature Importance (PFI) methodologies, thereby elucidating the relative importance of each contributory feature. In an unprecedented comparative analysis, the PSO-CatBoost model is rigorously benchmarked against six contemporary machine learning paradigms: CatBoost, XgBoost, AdaBoost, GBoost, Extra Trees, and Random Forest. Furthermore, this model is assessed against six empirical models for further comparison. The model exhibits superior predictive efficacy, evidenced by an exemplary coefficient of determination R-squared of 0.9847, surpassing the methodologies. This research introduces a new predictive model for CFRP-CC and represents a significant shift in concrete research, moving towards a more sophisticated, data-driven, and machine learning-focused methodology. This work thus establishes a new benchmark in the predictive modeling realm for CFRP-CC compressive strength, offering a robust and comprehensive analytical tool for both researchers and practitioners in the field. Lastly, a graphical user interface was designed for modeling the compressive strength of CFRP-CC to facilitate practical use.