REGRESSION-BASED MODELLING OF COMPRESSIVE STRENGTH OF RICE HUSK ASH–LATERITIC COMPRESSED EARTH BLOCKS USING LINEAR, INTERACTION, AND POLYNOMIAL REGRESSION MODELS

Abstract

incorporating rice husk ash (RHA) as a partial cement replacement. While previous studies have examined the effect of RHA on stabilised soils, limited research has applied regression-based modelling to quantify the combined effects of RHA content, stabiliser dosage, and curing age. A total of 18 mix combinations were prepared with RHA replacement levels of 0 - 40% and stabiliser contents of 6% and 10%. Compressive strength was evaluated at 7, 28, and 90 days in accordance with standard testing procedures using a universal testing machine. Linear, interaction, and polynomial regression models were developed, and model performance was assessed using R², adjusted R², RMSE, and MAPE. Results show that the interaction model achieved the best balance between accuracy and interpretability (R² = 0.91; adjusted R² = 0.89), while the polynomial model showed higher apparent accuracy (R² = 0.96) but signs of overfitting. Sensitivity analysis revealed that RHA exhibits a strong nonlinear influence on compressive strength, with a sensitivity index of 22.8 for the quadratic term. The study is limited by a relatively small dataset and the absence of external validation. However, the developed models provide a useful predictive framework for optimising stabilised lateritic blocks and reducing cement consumption.