<p>The present work introduces a novel approach to integrating graphical and animation components with a predictive analytics model for developing a web-based virtual laboratory application in Geotechnical Engineering, one of the core domains of Civil Engineering. Most virtual laboratories developed in this field so far are primarily simulation-based. In contrast, the current study utilises historical experimental data to train a machine learning model that predicts experimental outcomes. These predicted results dynamically influence the virtual experimentation process, thereby rendering the experimental experience and outcomes more realistic and enriching for students. As a proof of concept, an experiment titled Free Swell Index of Soils was developed as a smart virtual experiment. Experimental data were compiled from reputable research publications, the authors’ own data, academic projects and theses, and several machine learning regression models—namely Linear Regression, Support Vector Regression, Decision Tree Regression, and Random Forest Regression—were evaluated. Among these, the Random Forest Regression model demonstrated superior predictive performance and was therefore adopted for the final implementation.</p>