The compression index is a critical design parameter for estimating the consolidation behavior of clayey soils and planning construction schedules. However, its measurement typically requires more than a week for a single specimen. Therefore, to minimize reliance on consolidation tests, this study evaluated seven regression strategies and various combinations of readily available basic soil properties to identify optimal models balancing predictive accuracy and practical usability. A large database comprising more than 10,000 consolidation test results was compiled from diverse regions of South Korea, integrating parameters such as water content, void ratio, and Atterberg limits. The analysis revealed that dual-variable combinations, particularly those incorporating field-condition indicators and consistency limits, offered strong predictive performance. Conversely, using three or more variables resulted in diminishing returns, and it was not recommended unless accompanied by careful consideration of the data distribution and regression strategy. Among the regression strategies, k-nearest neighbor (KNN), artificial neural networks (ANNs), and linear regression (LR) demonstrated improved accuracy and reliability compared to traditional empirical formulas. The KNN model excelled when high-quality data were available, the ANN was robust under input variability, and LR provided a practical solution for preliminary assessments. Although some prediction errors remained—partly due to skewed variable distributions and the omission of potentially influential factors—the proposed models offer a reliable and scalable framework for estimating the compression index from basic geotechnical properties.
Yoo et al. (Tue,) studied this question.