Scholarlink

<p>Due to its low cost and its thermal and acoustic insulation properties, clay masonry is widely used in construction. Its compressive strength is the main mechanical property and is critical for structural design. This study aims to predict it using machine learning (ML) techniques. An experimental database was compiled from uniaxial compression tests on solid clay masonry specimens. First, the performance of 18 empirical models from the literature was evaluated. Then, several ML algorithms were developed, including least absolute shrinkage and selection operator regression, decision tree regression, support vector regression, bagging tree, gradient boosting, random forest regression, artificial neural networks, and Gaussian process regression (GPR). The models were trained on 80% of the data and tested on the remaining 20%, with hyperparameter optimisation and 10-fold cross-validation. The findings highlight the lower performance of traditional empirical models compared to ML methods. They also show the superior predictive ability of GPR over other ML algorithms for estimating the compressive strength of clay solid masonry. Sensitivity analysis confirms that masonry unit strength is the most influential factor, with notable nonlinear interactions involving mortar strength and geometric ratios, while joint thickness primarily acts as a regime modulator.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting the compressive strength of clay solid masonry using optimized machine learning algorithms and partial dependence plot analysis

  • 期刊论文

  • 发表日期：

    2026年2月1日

查看全文
  • Amar Messas,
  • Karim Benyahi,
  • Arezki Adjrad,
  • Youcef Bouafia,
  • Malika Belhocine

摘要

Due to its low cost and its thermal and acoustic insulation properties, clay masonry is widely used in construction. Its compressive strength is the main mechanical property and is critical for structural design. This study aims to predict it using machine learning (ML) techniques. An experimental database was compiled from uniaxial compression tests on solid clay masonry specimens. First, the performance of 18 empirical models from the literature was evaluated. Then, several ML algorithms were developed, including least absolute shrinkage and selection operator regression, decision tree regression, support vector regression, bagging tree, gradient boosting, random forest regression, artificial neural networks, and Gaussian process regression (GPR). The models were trained on 80% of the data and tested on the remaining 20%, with hyperparameter optimisation and 10-fold cross-validation. The findings highlight the lower performance of traditional empirical models compared to ML methods. They also show the superior predictive ability of GPR over other ML algorithms for estimating the compressive strength of clay solid masonry. Sensitivity analysis confirms that masonry unit strength is the most influential factor, with notable nonlinear interactions involving mortar strength and geometric ratios, while joint thickness primarily acts as a regime modulator.