When comparing the death burden across different regions only in the high-income region and the central Europe, Eastern Europe, and Central Asia region has the death burden trended down over the last three decades. This is despite the fact that pm25 exposure has decreased or stayed roughly level in almost all of the regions with the exception of South Asia seeing a reasonable in pm25 exposure. This suggests that while pm25 exposure likely plays a role in air pollution deaths there are other factors that are also incredibly important. One of the biggest pitfalls in this analysis was the lack of population data that likely played a big role in the death burden as more total deaths could result in more deaths by air pollution even if the number of deaths caused by pollution as a ratio remained the same.
Using the data from our data set we generated an ML model to predict the mean death burden for a given country in a given year. We then created a number of other models with certain columns dropped to see how important certain information was.
Based on the results we see that knowing the specific country impacted the model tremendously and losing the country-specific information drastically reduced the quality of the model. This may in part be due to a lack of population data so knowing the specific country may have been a proxy for that information. Knowing the specific country is also a proxy for knowing the GBD region so including the region information did not appear to add any additional information. It is also interesting to note that pm25 exposure information improved the model but the model still performed decently without that information.
While pm25 exposure is an important part of the air pollution death burden, its effect is not as pronounced as initially expected. This may in part be due to a lack of some data and it seems that specific country information acts as a proxy for a lot of the missing data. Ultimately the model could be improved upon with more data points as well as more descriptive data. It is also interesting to note that the exposure of pm25 has gone down over the last few decades in most regions and will hopefully continue to go down in the coming decades.