Machine Learning-Assisted Prediction and Multi-Objective Optimisation of Rice Husk Ash and Micro Fine Slag Stabilised Black Cotton Soil for Sustainable Pavement Subgrades

Abstract

Black cotton soil is a highly expansive geomaterial whose high plasticity, low soaked bearing capacity and moisture-driven volume change frequently cause pavement distress, subgrade deformation and serviceability loss. Sustainable soil stabilisation using rice husk ash (RHA) and Micro Fine Slag (MFS) is promising because it can combine agro-industrial residue utilisation with mechanical improvement of problematic expansive soils. This manuscript presents an integrated civil engineering and artificial intelligence/machine learning (AI/ML) framework for predicting and optimising the performance of RHA-MFS stabilised black cotton soil. The study is written as a validation-oriented predictive screening manuscript: modelled and augmented data are clearly separated from direct laboratory evidence, and final design adoption is made conditional on confirmatory experimental testing. Input features include RHA content, MFS content, curing age, liquid limit, plastic limit, plasticity index, maximum dry density, optimum moisture content, specific gravity and pH. Target responses include soaked California Bearing Ratio (CBR), unconfined compressive strength (UCS) and free swell index (FSI). Linear regression, ridge regression, random forest, Extra Trees, gradient boosting and support vector regression with radial basis function kernel (SVR-RBF) are benchmarked. Model assessment includes R2, RMSE, MAE, MAPE, cross-validation, residual diagnostics, feature importance, SHAP-style interpretation and prediction-uncertainty logic. The Results and Discussion section integrates civil engineering graphs and AI/ML diagnostic figures directly within the main text. The screening results support a practical validation domain of 10-15% RHA and 6-9% MFS at 28-56 days of curing, with 10% RHA + 6% MFS proposed as the first confirmatory laboratory candidate. The contribution of the paper is a reviewer-defensible methodology that uses AI/ML to reduce experimental search space while retaining geotechnical mechanism, material characterisation, sustainability reasoning and transparent limits on model-based claims.