Bayesian surrogate assisted neural network model to predict the hydrogen storage in 9-ethylcarbazole

• Deep neural network is applied to predict the hydrogen storage capacity. • Pearson, Spearman, and Kendal correlations are used to indicate the optimal parameter. • Time, pressure and temperature are found to be influential parameters. • Bayesian Surrogate Gaussian Process model predicted the hydrogen storage capacity accurately (R 2 =0.89). Optimization of the reaction conditions for hydrogen storage in 9-ethylcarbazole, an efficient liquid organic hydrogen carrier, is essential for advancing hydrogen energy applications. In this study, a deep neural network (DNN) model was developed to analyze the effects of key parameters such as temperature, initial pressure, catalyst type and dosage, and stirring speed on hydrogen storage capacity. Correlational analysis using Pearson, Spearman, and Kendal correlations identified time (r =0.83) as the most influential factor impacting the hydrogen storage positively, whereas catalyst dosage (r=-0.75) exhibited a strong negative correlation. Temperature (r=0.15) exhibited a weak relationship, suggesting nonlinear dependencies. To enhance the predictive accuracy, Bayesian Surrogate Random Forest (BSRF), Bayesian Surrogate Gaussian Process (BSGP), and Bayesian Surrogate Gradient Boost Regression Trees (BSGBRT) were integrated with deep neural networks (DNNs). Among these, the BSGP-DNN model exhibited the highest predictive performance, achieving an R 2 value of 0.89. The reliability of the model is further supported by a low Mean Absolute Error (0.0101), mean square error (0.0001), and relative deviation (3.2%). The error density curve centered around emphasized the model’s accuracy and uniform error distribution. This study employs a deep neural framework to analyze the effects of various reaction parameters on hydrogen storage capacity. Correlational analysis (using Pearson, Spearman, and Kendal correlations) is performed to find the optimal parameters as the inputs of the model. Moreover, we employed Bayesian Surrogate Random Forest (BSRF), Bayesian Surrogate Gaussian Process (BSGP), and Bayesian Surrogate Gradient Boost Regression Trees (BSGBRT) with deep neural networks (DNNs) for predicting hydrogen storage capacity (wt.%).