Interpretable Machine Learning Approaches for Assessing Maximum Force in Fiber-Reinforced Composites - Paper Abstract

Aug 2, 2026

Upload by Hazel

This paper investigates accurate prediction of the maximum force in fiber-reinforced composites using the CatBoost machine learning algorithm, focusing on interpretability for trustworthy decision support. Shapley additive explanations are integrated to quantify how each input variable affects model outputs at both local and global levels. The results show that feature importance derived from the model aligns with SHAP attributions, reinforcing the relevance of key parameters for interfacial-property-driven behavior. The study advances interpretable modeling for complex composite prediction tasks.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Introduction

Interpretable ML for maximum force prediction

Interpretable ML approach

SHAP methodology and feature attribution

Feature significance and error increase

Interpretable Machine Learning Approaches for Assessing Maximum Force in Fiber-Reinforced Composites Soheila Kookalani1, Erika Parn1, Ioannis Brilakis2 1Department of Engineering, University of Cambridge, Cambridge, UK. 2Laing O’Rourke Professor, Department of Engineering, University of Cambridge, Cambridge, UK. [sk2268@cam.ac.uk](sk2268@cam.ac.uk), [eap47@cam.ac.uk](eap47@cam.ac.uk), [ib340@cam.ac.uk](ib340@cam.ac.uk) Abstract This paper investigates the accurate prediction of the maximum force in fiber-reinforced composites using the CatBoost machine learning algorithm. The study incorporates the Shapley additive explanations technique to enhance interpretability, revealing the significance of the impact of each variable on the output at both local and global scales. The research demonstrates that Shapley additive explanations provides valuable insights into the decision-making process of the machine learning model, identifying influential variables for specific instances and contributing to a comprehensive understanding of the overall model predictions. Notably, the alignment between the feature importance analyses from the machine learning model and Shapley additive explanations reinforces the significance of certain parameters in predicting maximum force as an interfacial property. The study advances the prediction of interfacial properties in fiber-reinforced composites and underscores the value of interpretable machine learning method in offering insights into complex predictive models. Keywords – Machine learning; Interfacial properties; Maximum fore; Regression; Fiber-reinforced composites; Interpretability methods. 1 Introduction Fiber-reinforced composites have become integral materials in civil engineering applications, owing to their remarkable combination of stiffness, strength, and lightweight properties [1]–[3] . The mechanical performance of these composites is primarily dictated by their interfacial properties [4], [5] . the determination of interfacial properties through fiber pullout tests involves labor-intensive and time-consuming experimental and numerical methods. Hence, there is a pressing need for an accurate and efficient alternative for predicting interfacial properties, essential for the design and customization of composite materials. In recent years, machine learning (ML) techniques have emerged as promising substitutes for timeconsuming simulation processes that offer the advantage of low computational cost and high accuracy [6]–[10] . For instance, Mangalathu and Jeon [11] employed lasso regression for beam-column joints. Yao et al. [12] demonstrated the superiority of two-class support vector regression (SVR) over one-class SVR and logistic regression in mapping landslide susceptibility. Chopra et al. [13] investigated the efficiency of ML models such as decision trees (DT), random forests (RF), and neural networks in estimating concrete compressive strength. Their findings revealed the superior efficiency of the neural network model, followed by the RF method. Additionally, Das et al. [14] introduced a data-driven physics-informed approach for concrete crack estimation, showcasing the capability to predict infrastructure service life based on real-time monitoring data. In this study, the CatBoost algorithm is employed to predict the maximum force in fiber-reinforced composites. A grid search approach and K-fold crossvalidation are employed, utilizing a dataset comprising 922 samples to identify the optimum parameters of the ML model. Understanding why an ML model produces specific estimations and identifying the features influencing those estimations is crucial. Therefore, the Shapley additive explanations (SHAP) method is applied to comprehend the behavior of the ML model. The paper is organized as follows: Section 2 introduces SHAP as an interpretable ML approach; Section 3 presents a numerical example for maximum force prediction, and finally, Section 4 offers con

FAQs

What model is used to predict the maximum force in fiber-reinforced composites?

The study uses the CatBoost machine learning algorithm to predict maximum force from composite-related input variables.

How does SHAP improve interpretability in this research?

SHAP provides a systematic way to measure the impact of individual input features on predictions, showing effects at both local and global scales.

What do the authors find about the relationship between feature importance and SHAP explanations?

The feature importance analysis from CatBoost aligns with SHAP attributions, supporting the importance of certain parameters for predicting maximum force as an interfacial property.

Aug 2, 2026

Upload by Hazel