Title:
An Interpretable Machine Learning Framework for Thermal Conductivity Modelling of Fiber-Reinforced Cementitious Composites for Sustainable Building Insulation Applications (Prepublished)
Author(s):
K.A.P Wijesinghe, Madushan Rathnayaka, Chamila Gunasekara, David W. Law, Gamini Lanarolle, Hidallana-Gamage H.D, Lijing Wang
Publication:
Materials Journal
Volume:
Issue:
Appears on pages(s):
Keywords:
building insulation; data imputation; fiber-reinforced cement mortar; machine learning; thermal conductivity
DOI:
10.14359/51751900
Date:
8/7/2026
Abstract:
This study developed an interpretable machine learning (ML) framework to predict the thermal conductivity of fiber-reinforced mortars (FRMs). Traditionally, thermal conductivity is determined using experimental methods that are material-, time-, and cost-intensive, motivating the development of ML prediction models. Extreme Gradient Boosting (XGB), Random Forest (RF), Artificial Neural Networks (ANN), and Support Vector Machines (SVM) were evaluated as ML models. Missing data were addressed using Multiple Imputation by Chained Equations (MICE), K-Nearest Neighbors (KNN), and Singular Value Decomposition (SVD). The XGB model with SVD imputation achieved the best performance, with a test accuracy of 86%, which improved to 90% after feature selection and hyperparameter tuning. Model interpretation identified fiber diameter, fiber content, and sand proportion as key influencing factors. Experimental validation showed reliable predictions, with accuracy up to 97% for mid-range conductivity values. The proposed framework provides a practical tool for predicting the thermal conductivity of cementitious materials.