Application of RSM and ANN for Valorisation of Coffee Husk into Cellulose-Rich functional Ingredients

This work aimed to assess the use of artificial neural networks (ANNs) as an alternative tool for
modeling and predicting alkali pretreatment efficiency of coffee husk in comparison with
response surface methodology (RSM). Alkali concentration (NaOH, %), treatment time (min),
and temperature (°C) were selected as independent variables, and their effects on residual
cellulose concentration (%) were investigated. RSM was applied to develop a second-order
polynomial model, evaluate variable interactions, and identify statistically significant effects,
while ANN models were trained, validated, and tested to capture non-linear process behaviour.
The RSM analysis revealed significant quadratic effects of alkali concentration, time, and
temperature (p < 0.0001), with alkali concentration identified as the most influential parameter
affecting cellulose retention. ANN models demonstrated improved predictive performance over
RSM, exhibiting higher coefficients of determination (R²= 0.999) and lower root mean square
error (RMSE) values (0.068–0.141), indicating superior accuracy in modeling the complex
biomass–alkali interactions. Numerical optimization using the ANN model identified optimal
pretreatment conditions (≈2.55% NaOH, 105 min, and 68 °C), resulting in a maximum predicted
cellulose concentration of approximately 64%. The results confirm that ANNs are a suitable and
robust tool for optimizing and predicting alkali pretreatment outcomes in lignocellulosic biomass
processing. The optimized conditions resulted in a cellulose-rich residue suitable for downstream
enzymatic hydrolysis, facilitating the generation of value-added ingredients and supporting
sustainable coffee husk valorisation.