Beyond ANOVA: Esterification Optimization Code Repository

Companion code for:
Beyond ANOVA: A Structural Equation Modeling and Ensemble Machine Learning Approach to Batch Reactor Process Optimization

Author: Anfal Rababah
Preprint: ChemRxiv / Zenodo
DOI: https://doi.org/10.5281/zenodo.17623232
License: MIT / CC BY 4.0

Overview

This repository contains all analysis code used in the manuscript, including:

• Synthetic kinetic data generation (1,024-run factorial design)
• ANOVA with main effects and interaction terms
• Structural Equation Modeling (SEM) with mediation analysis
• Machine Learning (XGBoost) with SHAP interpretability
• Cross-method triangulation and factor-ranking comparison

The full pipeline enables transparent, reproducible reactor optimization using statistical, causal-modeling, and ML approaches.

---

System Requirements

Software Dependencies

• Python: 3.8+
• OS: Windows, macOS, Linux
• RAM: 8 GB minimum (16 GB recommended for ML)

Installation Time

~10–15 minutes depending on internet speed.

---

Installation

Option 1 — Using pip (Recommended)

# Clone the repository
cd Esterification_Optimization_Code

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Option 2 — Using conda
conda env create -f environment.yml
conda activate esterification-opt

Required Python Packages
numpy>=1.21.0
pandas>=1.3.0
scipy>=1.7.0
matplotlib>=3.4.0
seaborn>=0.11.0
statsmodels>=0.13.0
scikit-learn>=1.0.0
xgboost>=1.5.0
shap>=0.40.0
semopy>=2.3.0


(Complete list in requirements.txt)

## Usage Instructions

### Step 1 — Generate Dataset (~30 s)
cd 01_Data_Generation
python generate_kinetic_data.py


## Outputs

Data/esterification_data_1024.csv

Summary statistics

Step 2 — ANOVA (~2 min)
cd 02_ANOVA_Analysis
python anova_main_effects.py


Outputs

ANOVA table

Output/Figures/Figure2_ANOVA_MainEffects.png

Step 3 — SEM Analysis (~5 min)
cd 03_SEM_Analysis
python sem_model_comparison.py


Outputs

Fit indices

Path coefficients

Mediation effects

Figure3_SEM_PathDiagram.pdf


### Step 4 — ML Analysis (10–15 min)

cd 04_ML_Analysis
python xgboost_training.py
python shap_analysis.py
python partial_dependence_plots.py


Outputs

XGBoost model

SHAP summary plot

PDP plots

Feature rankings


### Step 5 — Cross-Method Comparison (~1 min)
cd 05_Cross_Method_Comparison
python triangulation_analysis.py

### Step 6 — Generate All Figures (~5 min)
cd 06_Visualizations
python generate_all_figures.py

Reproducing Specific Tables
python 02_ANOVA_Analysis/anova_main_effects.py --output-table   # Table 2
python 03_SEM_Analysis/sem_model_comparison.py --save-latex     # Table 3
python 04_ML_Analysis/model_comparison.py --export-csv          # Table 4
python 05_Cross_Method_Comparison/triangulation_analysis.py --format=latex  # Table 5



## Computational Performance

Hardware used

Intel Core i7-10700K

32 GB RAM

Windows 10 / Ubuntu 20.04


## Pipeline runtime

Data generation: 30 s

ANOVA: 2 min

SEM: 5 min

ML training: 15 min

Total: ~25 minutes


## Troubleshooting
- ModuleNotFoundError: semopy
pip install semopy

- XGBoost memory issues
Reduce CV folds:
cv_folds = 3
n_iter = 20

- Slow SHAP computation
Use TreeExplainer:

explainer = shap.TreeExplainer(model)

- Extending the Code
Use with Experimental Data

Replace synthetic generator with:

df = pd.read_csv('your_experimental_data.csv')


Expected columns:
['Temperature_C','Acid_Concentration_M','Catalyst_Concentration_M','Time_min','Yield_pct']

Add more ML models
models = {
    'XGBoost': xgb.XGBRegressor(),
    'Random Forest': RandomForestRegressor(),
    'Neural Network': MLPRegressor(),
    'Gradient Boosting': GradientBoostingRegressor()
}

## Citation

Please cite this work as:

@article{Rababah2025,
  title={Beyond ANOVA: A Structural Equation Modeling and Ensemble Machine Learning Approach to Batch Reactor Process Optimization},
  author={Anfal Rababah},
  journal={Zenodo},
  year={2025},
  doi={10.5281/zenodo.17623232}
}

## License

MIT License or CC BY 4.0.
Copyright (c) 2025 Anfal Rababah

## Contact
Email: Anfal0Rababah@gmail.com
GitHub Issues: Available in this repository
ResearchGate: Add your profile link

## Acknowledgments

NumPy, SciPy, pandas
Statsmodels
semopy
XGBoost + SHAP