A simulation of hierarchical self organizing maps to predict the compressive strength of fly ash-based geopolymers

Abstract

The complex, non-linear relationships between design parameters and compressive strength in fly ash-based geopolymer concretes (GPC) presents some challenges for classical empirical and statistical models. This study investigates the application of Hierarchical Self-Organizing Maps (HSOM) framework to model these non-linear interactions and predict compressive strength. In this study, 18 unique mix proportions were prepared and tested. Each mix proportion replicates 30 datasets, yielding a total of 540 compressive strength measurements. The dataset was used to train and evaluate the proposed HSOM against the standard Self-Organizing Maps (SOM) using a 5-fold cross validation strategy. The primary objective of this study is not to demonstrate superiority over established predictive models, rather, to demonstrate the feasibility and capability of the proposed HSOM framework as an alternative tool for GPC design. The findings demonstrate that HSOM achieves a coefficient of determination (R 2 ) of 0.80, outperforming the standard SOM (R 2 = 0.7). In terms of predictive accuracy, this outcome is also comparable to Response Surface Methodology. This study demonstrates that HSOM can provide complementary alternative for GPC mix design optimization especially where experimental data may be limited.