A Statistical and Machine Learning Approach to Engineering Properties of Heat-Treated Volcanic Rocks

Abstract This study investigates the impact of heat treatment on the engineering performance of volcanic rocks, with a particular focus on their physical, mechanical, and environmental durability characteristics. Eight different rock types from the İscehisar and Sandıklı regions of Afyonkarahisar, Turkey, were thermally treated at 1000 °C, 1055 °C, and 1150 °C. Laboratory analyses included porosity (PR), water absorption (WAP), sound speed propagation (SSP), uniaxial compressive strength (UCS), flexural strength (FS), and impact strength (IS), alongside environmental interaction assessments such as salt crystallization (SC), freeze–thaw (FT), and thermal shock (TS). Results revealed substantial microstructural changes—microcracking, feldspar recrystallization, and partial vitrification—confirmed through X-ray diffraction (XRD), scanning electron microscopy (SEM), differential thermal analysis (DTA)–thermogravimetric analysis (TG), and heating microscopy. These transformations led to increased porosity and WAP, and a notable reduction in mechanical strength, with analysis of variance (ANOVA) and Tukey's honest significant difference (HSD) indicating statistically significant differences (P < 0.001). Porosity rose by up to 40%, while UCS, FS, and IS declined by approximately 35%. To model these changes, 13 machine learning algorithms, including XGBoost, CatBoost, LightGBM, and Random Forest, were applied. The models achieved high predictive accuracy (R2 > 0.92), with SHAP (SHapley Additive exPlanations) analysis identifying WAP and UCS as dominant predictors. These findings emphasize the role of heat-induced microstructural changes in governing rock performance. The study offers a data-driven framework for optimizing volcanic rocks for sustainable use in construction, restoration, and architectural applications.