Semester of Graduation
Summer 2025
Degree
Master of Science in Civil Engineering (MSCE)
Department
Civil Engineering
Document Type
Thesis
Abstract
Protein rejection is a critical parameter in many applications of membrane separations because protein fouling can lead to reductions in both membrane permeability and selectivity. In the research described in this thesis, four machine learning (ML) models (Random Forest, XGBoost, Gradient Boosting, and Artificial Neural Networks) were leveraged to predict protein rejection rates during ultrafiltration and microfiltration with varied operational conditions, membrane properties, and protein concentrations. Using a dataset of 505 data points derived from 83 peer reviewed articles spanning 30 years, the models were trained and optimized through cross-validation and hyperparameter tuning to minimize mean squared error (MSE) and maximize the coefficient of determination (R2). The Gradient Boosting model was the most effective model, achieving an R2 of 0.69, which indicates its ability to explain 69% of the variance in unseen data, and an MSE of 0.033, demonstrating high prediction accuracy under varied operational conditions. Results demonstrated the superior predictive power of ensemble decision-tree ML models over traditional approaches and offer valuable insights into the design and operation of future membrane processes. The Shapley additive explanation method, which is rooted in cooperative game theory, was employed to reveal the impacts of different protein, operational, and membrane properties on model predictions. This analysis demonstrated that the ML models can effectively capture the significant effects of size exclusion in regulating membrane separation. By predicting the performance of yet-to-be-fabricated membranes, the findings have the potential to enhance industrial applications by guiding the creation of more effective and cost-efficient membrane configurations for water treatment.
Date
5-15-2025
Recommended Citation
Daniels, Gbenga E., "Predicting Protein Rejection In Membrane Separations Using Machine Learning Techniques" (2025). LSU Master's Theses. 6160.
https://repository.lsu.edu/gradschool_theses/6160
Committee Chair
Moe William