File size: 2,891 Bytes
b09c9f5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 |
---
language:
- en
tags:
- chemistry
- fuel
- engines
- YSI
---
# YSI Predictor β Yield Sooting Index Model
## π Overview
This repository contains a machine learning model for predicting the **Yield Sooting Index (YSI)** of single-component fuel molecules directly from their **SMILES** representation.
**YSI is a soot formation metric** used in combustion science.
- **Lower YSI β cleaner combustion**
- Highly relevant for **diesel replacement fuels**, **bio-fuels**, and **oxygenated fuels**.
This model supports:
- molecular design and optimization,
- genetic algorithms (e.g., CREM),
- Pareto optimization (CN vs YSI),
- rapid candidate screening.
---
## π§ How It Works
The prediction pipeline uses:
- **RDKit** β molecule parsing
- **Mordred** β 2D/3D molecular descriptors
- **FeatureSelector** β dimensionality reduction
- **Tree-based regression model** trained on experimental YSI values
**Prediction flow:**
1. Input SMILES β RDKit Molecule
2. Mordred descriptors generated
3. Feature selection applied
4. YSI predicted using trained regressor
Two model artifacts are included:
model.joblib # trained regressor
selector.joblib # feature selector used during training
---
## 𧬠Training Data
The model was trained using a curated dataset of **experimentally measured YSI values**, covering a diverse set of fuel molecule structures:
Includes:
- linear alkanes
- branched alkanes
- cyclic hydrocarbons
- aromatics
- oxygenated species (ethers, esters)
YSI range in dataset: **β 3 β 80**
---
## π Performance
Performance was evaluated on both training and **held-out test** sets.
### β Training Performance
| Metric | Score |
|--------|--------|
| RMSE | **6.9661** |
| MAE | **4.0581** |
| RΒ² | **0.9309** |
---
### π§ Test Performance
| Metric | Score |
|--------|--------|
| RMSE | **5.9667** |
| MAE | **3.8324** |
| RΒ² | **0.9440** |
| MAPE | **18.38%** |
The **test RΒ² = 0.9440** shows strong predictive accuracy.
---
### π Generalization Check
| Metric | Value |
|--------------|--------|
| Train RMSE | **6.9661** |
| Test RMSE | **5.9667** |
| Ξ (Test β Train) | **β0.9994** |
β‘οΈ The negative Ξ indicates **no overfitting**, and even **better test performance** due to more stable distribution.
---
## π Usage
Below is a minimal example showing how to use the model in Python.
> The feature calculation must match the training pipeline.
```python
import joblib
from rdkit import Chem
from shared_features import featurize_df, FeatureSelector
# Load model & selector
model = joblib.load("model.joblib")
selector = joblib.load("selector.joblib")
def predict_ysi(smiles: str):
mol = Chem.MolFromSmiles(smiles)
df = featurize_df([smiles])
X = selector.transform(df)
y = model.predict(X)
return float(y[0])
print(predict_ysi("CCCCCCC"))
|