Scholarlink

Concrete compressive strength refers to the material’s ability to withstand crushing or cracking under compression. It depends on factors like water-cement ratio, aggregate quality, and cement type. This study aims to understand concrete strength, ensure load-bearing capacity without failure, and explore the relationship between compressive strength (dependent variable) and other independent variables, including Cement, Blast Furnace Slag, Fly Ash, Water, Superplasticizer, Coarse Aggregate, Fine Aggregate, and Age. The dataset, sourced from Kaggle, contains 1030 instances and 9 attributes. It was split into multiple train-test ratios 60–40, 65–35, 70–30, 75–25, 80–20, and 85–15 for model training and evaluation using R-squared, Mean Absolute Error (MAE), and Mean Squared Error (MSE) metrics. Several machine learning algorithms were employed, such as Linear Regression, Elastic Net, Cat Boost, LightGBM, Gaussian Regression, Decision Tree, Random Forest, Bagging, Boosting, KNN, SVM, Neural Networks, Stacking Regressor, and GBoost with feature engineering. Among all models, Cat Boost outperformed others, achieving an R2 of 96%, MAE of 2%, and MSE of 11%. Linear Regression and Decision Tree also performed well but fell short of Cat Boost’s accuracy. The study’s novelty lies in comparing models across various splits and enhancing performance through feature engineering, offering practical insights for optimizing concrete mix design.

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 Concrete Compressive Strength Using Machine Learning: A Comparative Analysis of Models with Feature Engineering

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Peddi Praharshita

摘要

Concrete compressive strength refers to the material’s ability to withstand crushing or cracking under compression. It depends on factors like water-cement ratio, aggregate quality, and cement type. This study aims to understand concrete strength, ensure load-bearing capacity without failure, and explore the relationship between compressive strength (dependent variable) and other independent variables, including Cement, Blast Furnace Slag, Fly Ash, Water, Superplasticizer, Coarse Aggregate, Fine Aggregate, and Age. The dataset, sourced from Kaggle, contains 1030 instances and 9 attributes. It was split into multiple train-test ratios 60–40, 65–35, 70–30, 75–25, 80–20, and 85–15 for model training and evaluation using R-squared, Mean Absolute Error (MAE), and Mean Squared Error (MSE) metrics. Several machine learning algorithms were employed, such as Linear Regression, Elastic Net, Cat Boost, LightGBM, Gaussian Regression, Decision Tree, Random Forest, Bagging, Boosting, KNN, SVM, Neural Networks, Stacking Regressor, and GBoost with feature engineering. Among all models, Cat Boost outperformed others, achieving an R2 of 96%, MAE of 2%, and MSE of 11%. Linear Regression and Decision Tree also performed well but fell short of Cat Boost’s accuracy. The study’s novelty lies in comparing models across various splits and enhancing performance through feature engineering, offering practical insights for optimizing concrete mix design.