Regression Analysis of Deaths due to Environmental Factors

Summary

• Fuzzy-joining of various datasets on deaths due to environmental factors and global macroeconomic indicators
• Utilized Auto-ML (PyCaret) for regressor model selection and preliminary hyperparameter selection
• Hyperparamter tuning, auto-encoding, cross-validation, & ensembling of RandomForest, Extra Tree, and XGBoost

Context

First we take a look at a dataset for air pollutions deaths due over time for different countries. We visualize data and develop the linear regression models on the regional and national level. The linear regression models are valuable at predicting the general trend in deaths due over time and be used to predict future deaths, where past data appears to be highly linear. It has severe limitations for many countries. Polynomial models could be explored. To explore the interaction with other features, we proceed to merge different datasets between death due to environmental factors and socio-demographic/economic factors. We then examine them in 3 tree-based machine learning models.

Workflow

The set of scripts in the folder 'AirPollution_Scripts' are segmented for the purpose of clarity & convenience
The following is the suggested order for running the scripts:

• 'MLProject_AirPollution_Viz_LR.ipynb' - loading the initial Kaggle air pollution deaths dataset, processing, visualizing the base dataset, linear regression examination and future predictions.
• 'MLProject_DataProcessing_OECD_WB_GBD.ipynb' - loading raw data from world bank and GBD/OECD datasets, cleaning, and merging usable data. We test the use of fuzzy data set merging.
• 'MLProject_Exploratory_AutoML.ipynb' - loading the cleaning & merged dataset, model selection with PyCaret, and discussion of models/insights in modelling data.
• 'MLProject_ML_Models_SkLearn.ipynb' - development of selected machine learning models (Extra Trees, Random Forest, XGBoost), cross-validation, & Ensemble methods; discussion of model results

The aim is to explore the intial dataset (https://www.kaggle.com/datasets/akshat0giri/death-due-to-air-pollution-19902017?resource=download), supplement it, infer trends, and make predictions on future deaths due to environmental factors. The data is then supplement with:

• Global Burden of Disease Study 2019 Results via OECD.stats: https://stats.oecd.org/index.aspx?lang=en# (where the Kaggle dataset seems to have extracted and cleaned data from)