Dust_storm_modeling

Include spatial neighborhood to improve the prediction accuracy of a predictive model using XGBoost algorithm.

To create the environment and install the required packages:

For anaconda:

• open the anaconda promt
• cd to the location of the project
• run the Conda_Env_setup.bat file
• wait until the installation finish
• check the enviroment installation by typing:
  • conda activate Dstorm_311
  • conda list -n Dstorm_311

required packages:

• The code runs under Python version 3.11 with below packages
  • geopandas==0.14.1
  • rasterio==1.3.9
  • matplotlib==3.8.1
  • scikit-learn==1.3.2
  • numpy==1.26.2
  • pickles==0.1.1
  • pandas==2.1.3
  • xgboost==2.0.2
  • scipy==1.11.4

Python Files Decription

• Combo_randomForest.py / Combo_XGBoost.py

  • Random Search for tuning Spatial parameters (NAISS)

• Feature_Selection_RF.py / Feature_Selection_XGBoost.py

  • Dimentionality reduction and Feature Selection

• GWML_ANN.py / GWML_RandomForest.py / GWML_XGBoost.py

  • Binary Classification GWML

• GWML_RandomForest_Australia.py / GWML_XGBoost_Australia.py

  • Regression GWML

• RandomFOrest_Hypertunning_Main_Dataset.py / XGBOost_Hypertunning_Main_Dataset.py

  • Random Search for huper paramter tuning (NAISS)

• Test_bandwidth_GWML_RandomForest.py / Test_bandwidth_GWML_XGBoost.py

  • Bandwidth Exploration (NAISS)

• Test_HyperParameters_GWML_XGBoost.py / Test_HyperParameters_GWML_XGBoost.py

  • Hyper Parameter Exploration (NAISS)

• Test_Test_GWML_RandomForest.py / Test_Test_GWML_XGBoost.py

  • Data size Exploration (NAISS)

Analysis Visualizations

Bandwidth change effect on the accuracy and precision of the validation and test sets for Random Forest

left) SWML accuracy regarding bandwidth change. right) SWML precision regarding bandwidth change.

SWML local models’ accuracies for RF with Ds dataset

Bigger green circles indicate more accurate local models.

SWML elevation median importance for RF with Ds dataset

Elevation median was the most important feature in Global ML.

Impact of changing test dataset on the precision and accuracy for Random Forest

Upleft) Global ML accuracy for validation and test dataset. Upright) SWML accuracy for validation and test dataset. Bottomleft) Global ML precision for validation and test dataset. Bottomright) SWML precision for validation and test dataset.

Impact of changing hyperparameters on the precision and accuracy for Random Forest

Upleft) Global ML accuracy for validation and test dataset. Upright) SWML accuracy for validation and test dataset. Bottomleft) Global ML precision for validation and test dataset. Bottomright) SWML precision for validation and test dataset.