data4sustainability

Data science analysis for environmental sustainability. Rethinking baseline greenhouse gas emissions with data science.

Project Organization

├── LICENSE
├── Makefile           <- Makefile with commands like `make data` or `make train`
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
│                         the creator's initials, and a short `-` delimited description, e.g.
│                         `1.0-jqp-initial-data-exploration`.
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment (pip), e.g.
│                         generated with `pip freeze > requirements.txt`
│
├── environment.yml   <- The requirements file for reproducing the analysis environment (conda).
│
├── setup.py           <- makes project pip installable (pip install -e .) so src can be imported
├── src                <- Source code for use in this project.
│   ├── __init__.py    <- Makes src a Python module
│   │
│   ├── models         <- Scripts to train models and then use trained models to make
│   │   │                 predictions
│   │   ├── predict_model.py
│   │   └── train_model.py
│   │
│   └── visualization  <- Scripts to create exploratory and results oriented visualizations
│       └── visualize.py

Goal

Can you estimate the energy use intensity (EUI) of these new buildings if they were to have existed back in 2006 using historical data from buildings?

$$ \text{EUI} = \frac{\text{annual energy use (KBtu)}}{\text{building square footage (GSF)}} $$

Data Exploration

Pearson Correlations of Features with Energy Use Intensity (EUI)

Observations

• Location does not affect energy yse intensity (EUI)
• Year built and stories of a building are highly correlated with EUI
• EUI values range from 45 to 650

Challenges

• Small dataset (125 values after preprocessing)
• Limited feature set
• Some of energy gas, and oil usages provided are 0
  • Assume that data is invalid
• Extrapolating predictions to 2006

Modeling and Strategies

• Simple Linear Regression
  • Benefits: fast and easy to interpret
  • Disadvantages: often unrealistic of real-world data with noise
• Support Vector Regression
  • Benefits: shown to be very successful in practice
  • Disadvantages: can be computationally intensive and must search for optimal hyperparameters
• K-Nearest Neighbors
  • Benefits: fast, easy to interpret, and no training process
  • Disadvantages: non-parametric algorithm (model grows with dataset size)

Results

• Simple Linear Regression performs well predicting EUI with covariate, year built:

Metric	Value
$R^2$	0.26
MAE	41.34

Graph of Year Built vs EUI

Residual Plot

• Support Vector Regression performs well in predicting energy use intensity (EUI):

Metric	Value
$R^2$	-0.31 (arbitrarily worse than 0)
MAE	49.91

• K-Nearest Neighbors performs best in predicting energy use intensity (EUI):

Metric	Value
$R^2$	0.44
MAE	42.49

Predictions

Can be found in the submissions folder and the file called final_predictions.csv

Takeaways

• Promising performance for small subset of data and potentially erroneous samples
• With more data and features algorithms such as gradient boosting (XGBoost) and Neural Networks can be explored