Research Overview

This research investigates the application of advanced machine learning techniques for predicting oxygen transfer efficiency (OTE) in solid plunging jet aerators, with the broader objective of supporting efficient and sustainable wastewater treatment processes.

Research Objective

The study aims to develop and comparatively evaluate advanced machine learning models for predicting oxygen transfer efficiency under varying hydraulic and geometric conditions. The research also incorporates uncertainty quantification, statistical analysis and explainable artificial intelligence to evaluate model reliability and understand the influence of input parameters.

Methodology

An experimental dataset containing 320 observations was used for model development. The dataset was divided into 240 training observations and 80 independent testing observations. The predictor variables included:

Number of jet openings
Jet length
Disc flow area
Discharge

The response variable was oxygen transfer efficiency (OTE).

Machine Learning Models

Five predictive approaches were developed and compared:

Multiple Linear Regression (MLR)
Histogram-Based Gradient Boosting Regressor (HGBR)
Natural Gradient Boosting (NGBoost)
TabNet
Stacked Ensemble Model

The Stacked Ensemble incorporated multiple base learners and a meta-learner to improve predictive performance and generalization.

Model Evaluation

The models were evaluated using both deterministic and probabilistic performance measures, including:

R²
RMSE
MAE
NSE
PICP
MPIW

Additional analyses included residual analysis, Taylor diagrams, Friedman and Nemenyi statistical tests, uncertainty quantification and explainable AI-based sensitivity analysis.

Key Findings

The Stacked Ensemble model demonstrated the strongest overall predictive performance during testing, achieving R² = 0.9971, RMSE = 0.3399 and MAE = 0.1867. The study also found that discharge and aerator geometry were important parameters influencing oxygen transfer

efficiency.

Skills and Learning Outcomes

Through this research work, I gained exposure to:

Machine learning-based predictive modelling
Python-based data analysis and modelling
Environmental and wastewater engineering
Model performance evaluation
Uncertainty quantification
Statistical comparison of predictive models
Explainable artificial intelligence
Research paper preparation and technical writing

Publication Status

Manuscript submitted to a journal and currently under review.