Predicting design suitability of box-shaped sustainable timber structural members using machine learning and hyperparameter optimization

Abstract

The improvement of timber structures is essential for sustainable construction, focusing on enhancing ductility and energy dissipation. While machine learning offers a promising approach to accelerate and simplify the design process for such structures, its application in this field remains underexplored. This study develops digital models of box-shape timber members to assess their suitability for design (TS 647 Building Code for Timber Structures, Türk Standardlari Enstitüsü, Ankara, 1979), considering cross-section, span, loading, screws and material properties. A dataset of 2000 design cases was generated, incorporating variations in these factors based on design calculations in accordance with the relevant standards. After data preprocessing, machine learning (ML) classification algorithms, including Decision Tree (DT), Gaussian Naïve Bayes (GNB), K-Nearest Neighbour (KNN), Linear Discriminant Analysis Classification (LDA), Logistic Regression (LR), Support Vector Machine (SVM) and Voting Ensemble Classification (VEC), were employed to predict design suitability (Fx), where Fx indicates whether the generated design meets structural criteria. The study found that Fx positively correlated with cross-section, span and moment of inertia, while it had a strong negative correlation with moment, stress, shear and deflection. Accuracy scores ranged from 91.7% to 98.6%, with LR performing the best (98.6%), while GNB had the lowest score (91.7%). These results were supported by model performance metrics, namely precision, recall, F1, AUC and MCC scores. Additionally, hyperparameter optimization was applied to improve the performance metrics of the models, resulting in more accurate and reliable predictions. It improved DT, KNN and SVM, while LR, LDA and GNB showed no significant changes. These findings highlight the effectiveness of machine learning in predicting the suitability of timber structure designs, providing a more efficient approach to structural assessment. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2025.