Stacked machine learning framework for enhanced prediction of hydrogen sensing by nanocomposites: A multi-material approach

Nanocomposites with suitable structural and electronic properties have attracted significant attention as hydrogen-sensing materials. Despite conducting many laboratory-scale works, no generalized approach has been established to accurately predict the sensing behavior across different nanocomposites. This study develops a stacked machine learning (ML) framework that integrates artificial neural networks, adaptive neuro-fuzzy inference systems, group method of data handling, wavelet neural networks, and least-squares support vector regression to estimate the response of seventeen different nanocomposites toward hydrogen sensing. 527 experimental datasets under various environmental conditions were employed to train and validate the models. Unlike previous approaches restricted to single materials or small datasets, the proposed framework provides a more general and multi-material perspective. The stacked model achieved excellent accuracy (MAE = 1.19, RAE = 4.76 %, AARD = 6.65 %), clearly outperforming the individual base models. These results highlight the framework's strong generalization capability and its potential to accelerate the rational design and selection of nanocomposite-based hydrogen sensors. • Response of nanocomposite sensors controls by the H 2 concentration and temperature. • The most general model is developed for estimating sensor response for H 2 detection. • The model can accurately estimate the H 2 detection performance of 17 nanocomposites. • Designed model estimates testing group with the MAE = 1.73, RAE = 8.14 %, and r = 0.99214 • Temperature creates an upward/downward trend in sensor response toward H 2 detection.