Scholarlink

<p>The lateral confinement coefficient (K<sub>s</sub>) crucially influences the strength, ductility and seismic performances of the reinforced concrete columns (RCC), which is highly relevant for structural engineers and researchers. To address this purpose and enhance predictive performance this study utilizes novel hybrid evolutionary ensemble modeling approach by utilizing various input parameters from 204 data collected from different literatures. A typical random forest method provided less accuracy (R<sup>2</sup> = 0.84) in estimating the K<sub>s</sub> values, hence; Artificial Neural Network (ANN) based Random Forest (RF) and Gradient Boosting (GB) evolutionary models was employed in this study. Statistically, the performance of the ANN based evolutionary RF &amp; GB models (R<sup>2</sup> = 0.948 &amp; R<sup>2</sup> = 0.933) were comparatively greater than the typical RF ensemble model. Also, the performance of the ANN-RF and ANN-GB was compared by using various statistical parameters like RMSE for ANN-RF as 0.0943 and ANN-GB as 0.1065 along with other indices. In addition to the statistical indices, different visual representations like REC curve, Taylor diagram and residuals &amp; leverage plots were developed for comparing the performances of the model. The statistical values and plots reveal that both the models provide reliable and superior alternative for Ks prediction capable of handling complex non-linear relationships. Since there are more feature variables available for estimating the Ks value, the SHAP analysis and sensitivity analysis was also carried out to find out the feature variable which impacts the output while modifying the respective variables. Finally, the rank analysis was carried out by comparing all the performance indicators of both the models, which depicted that the ANN-RF model outperforms the ANN-GB model in training, whereas in testing conditions, ANN-GB model performed better for estimating the values of Ks, contributing safer design and resilient RCC columns. The Graphical User Interface (GUI) was developed based on the best performing model.</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

Hybrid ANN-ensemble models with interactive interface for predicting lateral confinement in RCC columns

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Ahmed Kateb Jumaah Al-Nussairi,
  • Yasser Taha Alzubaidi,
  • Ali K. Abdul Raheem,
  • Ahmed Adnan Hadi,
  • Ahmed A. Alamiery,
  • Aseel Smerat,
  • Mukhtar Hamid Abed,
  • Arun Kumar

摘要

The lateral confinement coefficient (Ks) crucially influences the strength, ductility and seismic performances of the reinforced concrete columns (RCC), which is highly relevant for structural engineers and researchers. To address this purpose and enhance predictive performance this study utilizes novel hybrid evolutionary ensemble modeling approach by utilizing various input parameters from 204 data collected from different literatures. A typical random forest method provided less accuracy (R2 = 0.84) in estimating the Ks values, hence; Artificial Neural Network (ANN) based Random Forest (RF) and Gradient Boosting (GB) evolutionary models was employed in this study. Statistically, the performance of the ANN based evolutionary RF & GB models (R2 = 0.948 & R2 = 0.933) were comparatively greater than the typical RF ensemble model. Also, the performance of the ANN-RF and ANN-GB was compared by using various statistical parameters like RMSE for ANN-RF as 0.0943 and ANN-GB as 0.1065 along with other indices. In addition to the statistical indices, different visual representations like REC curve, Taylor diagram and residuals & leverage plots were developed for comparing the performances of the model. The statistical values and plots reveal that both the models provide reliable and superior alternative for Ks prediction capable of handling complex non-linear relationships. Since there are more feature variables available for estimating the Ks value, the SHAP analysis and sensitivity analysis was also carried out to find out the feature variable which impacts the output while modifying the respective variables. Finally, the rank analysis was carried out by comparing all the performance indicators of both the models, which depicted that the ANN-RF model outperforms the ANN-GB model in training, whereas in testing conditions, ANN-GB model performed better for estimating the values of Ks, contributing safer design and resilient RCC columns. The Graphical User Interface (GUI) was developed based on the best performing model.