individual

What campus are you from?

Daytona Beach

Authors' Class Standing

Leah Oberkehr, Senior

Lead Presenter's Name

Leah Oberkehr

Faculty Mentor Name

Siddharth Parida

Abstract

This study presents a systematic machine learning-based framework to predict the compression index (Cc) and recompression index (Cr) of Florida soils using a database of 497 consolidation tests conducted for the Florida Department of Transportation. Traditional methods for estimating compressibility rely on time intensive laboratory tests or simplified empirical correlations that fail to capture the nonlinear, multivariate nature of soil behavior. To overcome these limitations, this research evaluates multiple predictive models—including polynomial regression, ridge regression, support vector regression (SVR), and random forests (RF)—to identify the most accurate and robust approach. Among these, the RF model showed the best performance, when applied to the database. Furthermore, a systematic feature selection process using Pearson correlation, distance correlation, and Gini importance identified initial void ratio, moisture content, dry unit weight, liquid limit, and plasticity index as the parameters that have the highest influence on model prediction. Feature reduction demonstrated that models using only these key parameters retained comparable predictive accuracy, enhancing efficiency for practical applications to sparse datasets. Furthermore, the study explored the effects of soil classification on model performance, comparing conventional Unified Soil Classification System (USCS) groupings with unsupervised K-means clustering. Data-driven clustering produced more homogeneous subsets, improving model robustness and mitigating the poor performance observed in sparse coarse-grained USCS categories. The proposed framework highlights the potential of machine learning—especially RF models combined with intelligent feature selection and clustering—to develop interpretable, location-specific, and highly accurate predictive tools for geotechnical engineering applications, reducing reliance on time-consuming laboratory testing.

Did this research project receive funding support from the Office of Undergraduate Research.

No

Share

COinS
 

A Machine Learning Based Systematic Framework for Modelling of Compression and Recompression Indices for Florida Soils

This study presents a systematic machine learning-based framework to predict the compression index (Cc) and recompression index (Cr) of Florida soils using a database of 497 consolidation tests conducted for the Florida Department of Transportation. Traditional methods for estimating compressibility rely on time intensive laboratory tests or simplified empirical correlations that fail to capture the nonlinear, multivariate nature of soil behavior. To overcome these limitations, this research evaluates multiple predictive models—including polynomial regression, ridge regression, support vector regression (SVR), and random forests (RF)—to identify the most accurate and robust approach. Among these, the RF model showed the best performance, when applied to the database. Furthermore, a systematic feature selection process using Pearson correlation, distance correlation, and Gini importance identified initial void ratio, moisture content, dry unit weight, liquid limit, and plasticity index as the parameters that have the highest influence on model prediction. Feature reduction demonstrated that models using only these key parameters retained comparable predictive accuracy, enhancing efficiency for practical applications to sparse datasets. Furthermore, the study explored the effects of soil classification on model performance, comparing conventional Unified Soil Classification System (USCS) groupings with unsupervised K-means clustering. Data-driven clustering produced more homogeneous subsets, improving model robustness and mitigating the poor performance observed in sparse coarse-grained USCS categories. The proposed framework highlights the potential of machine learning—especially RF models combined with intelligent feature selection and clustering—to develop interpretable, location-specific, and highly accurate predictive tools for geotechnical engineering applications, reducing reliance on time-consuming laboratory testing.

 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.