Final Project for Linear Regression

Contributors:

• Wayne Lam
• Omar Hussain

Notebook: RegressionProject.ipynb

Data Sources:

• Suicide rates by country and year from: http://ghdx.healthdata.org/gbd-results-tool
• UN WHR from Kaggle by user: Alcides Oxa
• Capitals and lat,long from: https://www.jasom.net/list-of-capital-cities-with-latitude-and-longitude/
• Mental Health Disorders from: https://github.com/Botafogo1894/Project2/blob/master/mental_disorder_substance_use.csv
• Vector data for countries from: https://www.naturalearthdata.com/downloads/110m-cultural-vectors/

EDA and Data Visualizations:

• Relevent Features: Alcohol Interactions Scatterplots
• Less Relevent Features: Certain Happiness Metrics
• World Suicide Rates (mean over 2005-2016)

Feature Engineering:

• Log GDP/Capita (already in UN WHR) and Log Suicide Rate
• Feature Interactions with R^2 > 0.1 (13 out of 24 are alcohol dependency related)

Feature Selection and Model Selection:

• Wrapper Method (Recursive Feature Selection) to select features
• Embedded Method (Lasso Regression) to select features
• GridSearch for hyperparameter tuning for regularized regressions
• Ridge Regression generally had the highest R^2 score and lowest RMSE: 0.628, 4.211 respectively with mean 10.459, std 7.274