HTL Bio-Oil Yield Prediction using Machine Learning

---

Project Overview

This project develops a modular machine learning framework for predicting bio-oil yield from Hydrothermal Liquefaction (HTL) experiments.

Unlike notebook-based implementations, the project provides a reusable software architecture for preprocessing, training, hyperparameter optimization, explainability, and visualization.

---

Key Features

• Modular ML framework
• Six regression algorithms
• Hyperparameter optimization
• SHAP explainability
• Partial Dependence Plots
• Permutation Importance
• Model comparison dashboard
• Automated experiment logging
• Publication-quality visualizations

---

Dataset

Property	Value
Samples	2284
Features	29
Task	Regression
Target	Bio-Oil Yield (%)

---

Workflow

Dataset
      │
      ▼
Preprocessing
      │
      ▼
Model Training
      │
      ▼
Hyperparameter Optimization
      │
      ▼
Explainability
      │
      ▼
Dashboard

---

Results

Model	Test R²	Test MAE
🥇 Tuned XGBoost	0.8689	4.2589
Tuned Random Forest	0.8659	4.2796
XGBoost	0.8634	4.4094
Random Forest	0.8624	4.4111
Extra Trees	0.8619	4.1621
CatBoost	0.8370	5.1441

---

Explainability

Top predictive features:

• Lipids
• Temperature
• Higher Heating Value
• Proteins
• Fatty Acids

---

Repository Structure

src/
 ├── core/
 ├── models/
 ├── experiments/
 ├── visualization/
 └── legacy/

---

Installation

pip install -r requirements.txt

---

Run

python -m src.models.xgboost

python -m src.visualization.model_dashboard

---

Future Work

• Bayesian Optimization
• LightGBM
• Deep Learning
• Streamlit Deployment
• Multi-output HTL prediction

---

Author

Rohan

Indian Institute of Technology Kharagpur