A powerful, model-agnostic tool to measure spatially varying uncertainty of machine learning models. GeoConfromal is an extension of Conformal prediction.
GeoConformal, in theory, supports any machine learning model with spatial (e.g., coordinates) and aspatial (e.g., area of the living) features as the input.
from GeoConformalPrediction import GeoConformalRegressor
from xgboost import XGBRegressor
from sklearn.model_selection import train_test_split
X_train, X_temp, y_train, y_temp, loc_train, loc_temp = train_test_split(X, y, loc, train_size=0.8, random_state=42)
X_calib, X_test, y_calib, y_test, loc_calib, loc_test = train_test_split(X_temp, y_temp, loc_temp, train_size=0.5, random_state=42)
model = XGBRegressor(n_estimators=500, max_depth=3, min_child_weight=1.0, colsample_bytree=1.0).fit(X_train.values, y_train.values)
geocp_regressoer = GeoConformalRegressor(predict_f=model.predict, x_calib=X_calib.values, y_calib=y_calib.values, coord_calib=loc_calib.values, bandwidth=0.15, miscoverage_level=0.1)
results = geocp_regressoer.geo_conformalize(X_test.values, y_test.values, loc_test.values)from GeoConformalPrediction import GeoConformalClassifier
from xgboost import XGBClassifier
from sklearn.model_selection import train_test_split
X_train, X_temp, y_train, y_temp, loc_train, loc_temp = train_test_split(X, y, loc, train_size=0.8, random_state=42)
X_calib, X_test, y_calib, y_test, loc_calib, loc_test = train_test_split(X_temp, y_temp, loc_temp, train_size=0.5, random_state=42)
model = XGBClassifier(n_estimators=100, max_depth=2, min_child_weight=1.0, colsample_bytree=1.0).fit(X_train, y_train)
geocp_classifier = GeoConformalClassifier(predict_f=model.predict_proba, x_calib=X_calib.values, y_calib=y_calib, coord_calib=loc_calib.values, bandwidth=0.2, miscoverage_level=0.1, nonconformity_score='aps')
results = geocp_classifier.geo_conformalize(X_test.values, y_test.values, loc_test.values)This repository hosts the code base for the paper
GeoConformal prediction: a model-agnostic framework of measuring the uncertainty of spatial prediction
Xiayin Lou, Peng Luo, Liqiu Meng
Annals of the American Association of Geographers
Link to Paper
If you find this work useful, please consider cite:
@article{lou2025geoconformal,
title={GeoConformal Prediction: a model-agnostic framework for measuring the uncertainty of spatial prediction},
author={Lou, Xiayin and Luo, Peng and Meng, Liqiu},
journal={Annals of the American Association of Geographers},
pages={1--28},
year={2025},
publisher={Taylor \& Francis}
}