📢 Publish Your Research for Free - Full APC Waiver, No Hidden Charges. Submit Your Article Today! Submit Now →
JImage

ARTICLE TYPE : RESEARCH ARTICLE

Published on :   14 Nov 2026, Volume - 2
Journal Title :   WebLog Journal of Chemical Engineering | WebLog J Chem Eng | WJCHE
Source URL:   weblog icon https://weblogoa.com/articles/wjche.2026.k1401
Permanent Identifier (DOI) :   doi icon https://doi.org/10.5281/zenodo.24271067

Comparative Evaluation of Response Surface Methodology (RSM) and Artificial Neural Networks (ANN) for Alumina Extraction from Calcined Owhe Kaolinite Using HNO3 / H2 O2 Leaching

Mabel Keke 1 *
Ikechukwu A. Nnanwube 2
Okechukwu D. Onukwuli 3
Samson O. Okpo 1
1Chemical Engineering Department, Southern Delta University Ozoro, Delta State, Nigeria
2Chemical Engineering Department, Michael Okpara University of Agriculture, Umudike, Nigeria
3Chemical Engineering Department, Nnamdi Azikiwe University, Awka, Nigeria

Abstract

Alumina (Al2 O3 ) is an important industrial material used in ceramics, refractories, catalysts, and high-purity applications. This study investigated alumina extraction from calcined Owhe kaolinite, Delta State, Nigeria, using HNO3 /H2 O2 leaching and compared Response Surface Methodology (RSM), Artificial Neural Network (ANN), and Gaussian Process Regression (GPR) for process modelling and prediction. Five operating variables leaching temperature, HNO3 concentration, stirring rate, leaching time, and H2 O2 concentration were investigated using 32-run rotatable half fraction Central Composite Design. The quadratic RSM model showed a strong fit to experimental data (R2 = 0.997, adjusted R2 = 0.991, predicted R2 = 0.919; F = 176.9, p <0.0001), although the significant lack of fit (p = 0.00044) indicated residual systematic variation. Under a common leave one-out cross-validation framework, RSM provided superior predictive performance (R2 = 0.919, RMSE = 2.22%) compared with nested-CV ANN (R2 = 0.749, RMSE = 3.89%) and GPR (R2 = 0.646, RMSE = 4.62%). Numerical optimisation using RSM predicted a maximum alumina recovery of 89.5% at 85.1°C, 6.18 M HNO3 , 469 rpm, 108 min, and 5.96 M H2 O2 , with a 95% single-run prediction interval of approximately ±1.8 percentage points. Replicate experiments at the previously identified optimum yielded 86.37% recovery, confirming experimental reproducibility but not directly validating the revised optimum. Overall, quadratic RSM outperformed ANN and GPR in predicting the optimum for the present dataset, although further experimental validation remains necessary.

Keywords: Leaching; HNO3 ; H2 O2 ; Alumina; Optimization; Dissolution; Metakaolin

Citation

Keke M, Nnanwube IA, Onukwuli OD, Okpo SO. Comparative Evaluation of Response Surface Methodology (RSM) and Artificial Neural Networks (ANN) for Alumina Extraction from Calcined Owhe Kaolinite Using HNO3 /H2 O2 Leaching. WebLog J Chem Eng. wjche.2026. k1401. https://doi.org/10.5281/zenodo.24271067