Life Expectancy Project

Team Members

Brayton Hall, Sailaja Karra

• Data
• EDA
• Model
• Conclusions

Project Goals

Our aim was to determine the most important features for anticipating a country's life expectancy. Many features are economic in nature, and determining the most predictive ones may inform more effective resource allocation to increase life expectancy.

Data Collection

Our data was collected via Kaggle from the World Health Organization's Life Expectancy dataset under its Global Health Observatory (GHO). Important features among 22 initial independent variables include:

• schooling
• adult mortality rate
• bmi
• income index
• HIV/AIDS ratio
• infant mortality
• gdp
• population

EDA

Data Cleaning

We chose to drop 'Hepatitis B' due to missing values, as well as Country and Year because they were dominating other predictors. We then imputed median for 'schooling', 'alcohol', 'GDP', and all economic features missing values. We turned our only binary categorical variable 'Status', into 0 or 1 for 'Developing' or 'Developed', and dropped all remaining missing values.

Data Exploration

We noticed immediate strong correlations between the following features and our target variable (lifex): schooling .78, adult_mort -.67, bmi .59, status .51. Scatterplots showed a strong linear relationship with lifex.

Model & Results

We used recursive feature engineering with cross validation, linear regression, Lasso L1, Ridge L2, and GridSearchCV to produce our best model : Ridge L2 (alpha: .01) with a root mean squared error of 3.69, meaning it is, on average, 3.69 years off when predicting the true values.

Regression Analysis

Recursive feature elimination revealed the most important features in our model, with Income (Composition of Resources) as the main driving factor for life expectancy, followed by Schooling and HIV/AIDS. Our residuals from the model are normally distributed and symmetric, indicating that the assumptions of linearity and homoscedasticity are met.

Conclusions

From our Ridge L2 model, using recursive feature selection, we concluded that the most important features for determining a country's life expectancy are Income (Composition of Resources), Schooling, and HIV/AIDS, though it would be incorrect to conclude that improving these factors automatically increases life expectancy. These features are mostly likely pointers for deeper factors in countries that are more causally related to Life Expectancy, which further research could identify in order to make effective public policy decisions.