Scholarlink

<p>The approach for determining the mix design of bituminous mix in India as set forth by MORTH relies on the Marshall method. However, the traditional Marshall mix design method requires determination of several parameters leading to significant time consumption, high costs, complexity and the need for skilled personnel. This study employs innovative machine learning algorithms i.e. artificial neural network (ANN), support vector machine (SVM) and Gaussian process (GP) based on kernel functions to develop prediction models for Marshall stability (MS) of modified bituminous mixes. A total of 51 experimental data points were collected, with five independent parameters namely bitumen penetration (BP), optimum bitumen content (OBC), type of anti-stripping agent (ASA type), percentage of anti-stripping agent (ASA %) and percentage of polypropylene fibre (PPF %) selected as predictor variables for predicting MS. The performance of ML models was assessed using various statistical metrics. Among the models, the GP with normalized polynomial kernel function (NPKF) achieved superior predictive accuracy with R values of 0.952 and 0.9613 and low RMSE values of 0.514 and 1.1356 in training and testing, respectively. Sensitivity analysis revealed that OBC was the most influential factor governing MS. The findings demonstrate that ML based prediction can serve as a reliable, time-efficient, and cost-effective alternative to traditional laboratory testing, offering practical benefits for pavement engineers in mix design optimization and quality control.</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

Prediction models for Marshall stability using ANN, SVM andGP machine learning algorithms: a comparative study

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Samrity Jalota,
  • Manju Suthar

摘要

The approach for determining the mix design of bituminous mix in India as set forth by MORTH relies on the Marshall method. However, the traditional Marshall mix design method requires determination of several parameters leading to significant time consumption, high costs, complexity and the need for skilled personnel. This study employs innovative machine learning algorithms i.e. artificial neural network (ANN), support vector machine (SVM) and Gaussian process (GP) based on kernel functions to develop prediction models for Marshall stability (MS) of modified bituminous mixes. A total of 51 experimental data points were collected, with five independent parameters namely bitumen penetration (BP), optimum bitumen content (OBC), type of anti-stripping agent (ASA type), percentage of anti-stripping agent (ASA %) and percentage of polypropylene fibre (PPF %) selected as predictor variables for predicting MS. The performance of ML models was assessed using various statistical metrics. Among the models, the GP with normalized polynomial kernel function (NPKF) achieved superior predictive accuracy with R values of 0.952 and 0.9613 and low RMSE values of 0.514 and 1.1356 in training and testing, respectively. Sensitivity analysis revealed that OBC was the most influential factor governing MS. The findings demonstrate that ML based prediction can serve as a reliable, time-efficient, and cost-effective alternative to traditional laboratory testing, offering practical benefits for pavement engineers in mix design optimization and quality control.