This project aims to detect biases in open building datasets. It includes a suite of Python scripts designed to analyze building dataset accuracy, visualize data through barplots, and conduct correlation and regression analysis.
barplots.pybuildings_accuracy.pycorrelation_regression.py
This script conducts an accuracy analysis on open building datasets.
Inputs:
- At least one building dataset for analysis.
- A reference building dataset (ground-truth).
- A file specifying study areas, which includes sensitive variables.
Features:
- Calculates true positives, false positives, and false negatives using the IoU method (default threshold: 0.5, adjustable).
- Outputs Excel files for each building dataset at both building and study area levels.
Customization:
- Specify pathfiles to your datasets.
- Adjust column names for sensitive variables in the study area file in the 'append results' section.
This script generates barplots based on the Excel files created by buildings_accuracy.py.
Input:
- Excel files on tile level from
buildings_accuracy.py.
Output:
- Barplots comparing each sensitive variable to the false negative rate (or other fairness metrics) across all building datasets.
- Equality of opportunity calculation for each variable.
Performs correlation analysis and weighted linear regression.
Input:
- Excel file on tile level from
buildings_accuracy.py.
Features:
- Conducts correlation analysis between sensitive variables and the false negative rate (or other metrics).
- Performs weighted linear regression (default weight: building density, adjustable).
- Prints correlation results and displays significant linear regression results through scatter plots.