Case Studies of Real World ML Implementation

Click on an image below to read the case study project details.  Each case study is organized by the steps used in the ML Toolkit framework/workflow:  1) Define Problem, 2) Get Data, 3) Prepare Data, 4) Train Model, and 5) Deploy Model. Each case study includes a link to access the project code via a Google Colab notebook.  Each case study also includes a short project overview video as well as a code walk-through video.

Carbon Diversion: High-rate Contact Stabilization

Keywords: Disinfection, Lab data, Sensor data, Mechanistic model, Transformation, KDE, Cross-validation, XGBoost

Paracetic Acid (PAA) Disinfection Dosage Control

Keywords: Struvite, Nutrient recovery, Lab data, Sensor data, Cross-validation, XGBoost, Dimension reduction

Performance Prediction and Setpoint Adjustment of Struvite Precipitation Reactor

Keywords: A-stage, Supervised learning, Multivariate imputation, Linear imputation, Cross-validation, XGBoost

Total Solids Prediction: Acoustic Sensing and ML for Holistic Biosolids Optimization

Keywords: A-stage process, Offline-decision support tool, Multivariate data imputation, Artificial Contrasts with Ensembles (ACE), Wrapper Feature Selection, Supervised learning, Support Vector Machines (SVM), Decision tree ensembles, Data imbalance treatment, Shapley Additive exPlanations (SHAP)

Keywords: Ammonia-Based Aeration Control (ABAC), Ammonia versus NOx (AvN, partial denitrification anammox (PdNA), Hybrid model, XGBoost, LSTM, Fast fourier transform

Biological Nutrient Removal: ABAC / AvN + PdNA