Maskininlärning vs. Ekonomiska modeller : En jämförelse av faktorer som påverkar Bruttonationalprodukten i Sverige

It is widely discussed how machine learning could be used to improve the economy. It was therefore identified that the predictability of gross domestic product (GDP) per capita using other data is a problem well suited for machine learning. However, there are already other systems and experts that can do these calculations, thus the question ’Which features are identifiable as important via SHAP analysis on an XGBoost model that predicts the GDP of Sweden, and are these features what the Swedish Ministry of Finance deems as important factors when analyzing Sweden’s GDP?’ was born. Two machine learning models were made through the XGBoost Python library, one with only the factors deemed important by the paper published in 2023 from the Swedish Ministry of Finance and one with all factors deemed important by previous research, including the ones identified by the Swedish Ministry of Finance. It was found that the model with more factors was slightly more accurate, but more interestingly, it did not consider the important factors from the model based on the Swedish Ministry of Science important at all. It was found that in the full model, the most important features were the gender education parity index, population, daily caloric supply, and CO2 emissions. Thus, the results suggest that this type of machine learning model does not entirely coincide with the findings by the Swedish Ministry of Finance. Also, there are some sources of error, mainly related to the data available and data recency.