The aim of this project is to determine most influential features regarding the damage suffered by houses caused by forest fires using multivariate statistical models.
This research was supported by Fondecyt grant 1191543, Integration of remote sensing and direct data for multi-scale, dynamic mapping of urban exposure to flood, earthquakes and fire hazards (P. Aguirre).
- Exploratory Data Analysis
- Data Cleaning
- Feature Selection
- Binary Classification (
LightGBM) - Feature Importance (
shap)
Clone this repository, move to the folder and run on your favourite environment ('conda', 'mamba', 'venv', 'docker', etc.) the following:
python -m pip install -e wfm
The flag -e mean this is a installation in developing mode, in order you can modify some parameters.
You must have an input folder where each scenario must be another folder with georeferenced files inside, e.f. .shp.
For exploratory data analysis you can use the file eda_cli.py as follow
python eda_cli.py --input_path {YOUR_INPUT_PATH} --output_path {YOUR_OUTPUT_PATH}
Both arguments are optionals, for default input and output paths are input and exploratory_data_analysis respectively.
Same for data modeling using main.py as follow
python main.py --input_path {YOUR_INPUT_PATH} --output_path {YOUR_OUTPUT_PATH}
In this case, default values are 'input' and 'output' respectively.