Resilient Modulus of SCBA-Modified Granular Subbases: Random Experimental Forest Analysis

Determining the resilient modulus of granular subbases modified with agro-industrial residues is experimentally demanding and subject to substantial variability. This study evaluates the axial resilient response of a granular subbase modified with sugarcane bagasse ash (SCBA) and examines the ability of a Random Forest model to estimate the resilient modulus under different loading and confinement conditions. SCBA was incorporated as an addition relative to the dry mass of the granular material. The resilient-modulus analysis considered SCBA contents of 0%, 5%, and 10%. Furthermore, a 15% mixture was additionally characterized through compaction and the California Bearing Ratio (CBR) tests but was not included in the quantitative resilient-modulus analysis because a sufficiently reliable response was not obtained under the applied cyclic stress sequence. Among the resilient-modulus conditions analyzed, the 5% mixture exhibited the most favorable mechanical response, with greater stability and lower permanent strain accumulation than the control and 10% mixtures; however, this value should not be interpreted as a global optimum because intermediate dosages around 5% were not evaluated. For the machine learning analysis, the dependence among consecutive experimental readings was explicitly considered and variables derived from the resilient cycle were excluded to prevent target leakage. Under the primary leave-one-sequence-out validation scheme, the Random Forest model using variables known before testing achieved R2 = 0.78 (95% block-bootstrap CI 0.59–0.89) and RMSE = 38.9 MPa, comparable to the classical k-θ power-law model (R2 = 0.84; RMSE = 32.5 MPa). Extrapolation to an unseen SCBA level remained limited. These results indicate that Random Forest can serve as a complementary surrogate within the tested stress and material domain, but additional specimens and intermediate SCBA dosages are required before generalizing the model or defining an optimum ash content.