Statistical Modeling for the Prediction of California Bearing Ratio Values From Compaction Characteristics and Some Index Properties of Soils: A Case Study of Seka Town, Ethiopia

The California bearing ratio (CBR) is a key parameter for flexible pavement and airfield runway design; however, its laboratory determination is labor‐intensive, time‐consuming, and costly. This study developed a predictive model for soaked CBR values using compaction characteristics and some index properties of soil to provide a cost‐effective estimation tool for subgrade soils in Seka Town, Ethiopia. A total of 39 fine‐grained soil samples were collected from an ongoing roadway construction project. Thirty samples were used for model calibration, while the remaining nine samples were reserved for internal validation. Laboratory testing was conducted in accordance with ASTM and American Association of State Highway and Transportation Officials (AASHTO) standards to determine index properties, including percentage passing sieve No. 200, liquid limit (LL), plastic limit (PL), and plasticity index (PI) as well as compaction characteristics such as maximum dry density (MDD) and optimum moisture content (OMC) and CBR values. The investigated soils were predominantly classified as MH and ML under the Unified Soil Classification System (USCS) and A‐7‐5 and A‐7‐6 under the AASHTO classification system. Simple and multiple linear regression (MLR) analyses were performed using SPSS software. Among the developed models, the MLR equation, CBR = −12.124 − 0.077LL − 0.178OMC + 20.37MDD, demonstrated a strong correlation (adjusted R ²  = 0.887) and the best predictive performance with low error metrics (MAE = 0.511, MAPE = 6.84%, and RMSE = 0.66). Internal validation confirmed good predictive accuracy, with an average variation of 3.64%. However, external validation using a secondary dataset resulted in a significant average variation with a low correlation ( R ²  = 0.381) and relatively high error metrics (MAE = 4.471, MAPE = 43.65% and RMSE = 4.76), indicating poor generalizability across different geographic regions. Therefore, the developed model should be considered preliminary and locally applicable to soils with characteristics similar to those of the study area. Further validation using larger, multiregion datasets and advanced modeling approaches is recommended before broader practical engineering application.