Machine learning assisted chitin Extraction: Source and yield prediction from crustacean biomass using deep eutectic solvents

Abstract

Chitin is the primary polymer after cellulose found in the crustacean shells, insects, and fungi. In recent years, extracting chitin from crustacean shells has garnered attention as a means to convert waste into valuable resources, given its promising applications in various fields. While several methods have been employed to recover chitin, emerging green solvents such as deep eutectic solvents (DES) are increasingly favored to address the challenges posed by traditional chemical, enzymatic, and biological approaches. Chitin extraction efficiency is governed by multiple interdependent factors with complex, non-linear interactions, making conventional trial-and-error optimization both time-consuming and inefficient. These challenges highlight the need for advanced approaches, such as machine learning, to accurately model and optimize the process. Recognized for its reliability and adaptability, machine learning offers an effective approach to modeling the complex, non-linear interactions among process variables in chitin extraction, thereby enabling precise yield prediction and process optimization. The present work proposes a machine learning approach designed to accurately estimate chitin yield and optimize the key parameters governing its extraction. Experimental datasets from chitin extraction trials were used to train and evaluate multiple ML algorithms. Key process variables were treated as input features, with chitin yield as the output target. Among the models tested, XGBoost and Decision Tree achieved the highest predictive accuracy (R² = 0.99) with minimal root mean square error (RMSE) and mean absolute error (MAE) values. Model-driven optimization was then employed to identify the most favorable combination of process variables. Experimental extractions conducted under the predicted optimal conditions produced chitin yields closely matching the model forecasts, confirming high prediction accuracy. This research demonstrates that machine learning can serve as a cost-effective, efficient method for predicting chitin yield and paves the way for addressing complex issues involving large datasets. Further, the method offers a scalable strategy for enhancing biopolymer recovery and overcoming the multifaceted challenges posed by extensive datasets in sustainable materials research.