University homework project where I explored what factors most influence happiness scores across 156 countries. Covered correlation analysis, distribution checks, and basic predictive modelling.
Industry
B2B Services
Scope
Web
Duration
6 weeks
Stage
Scale-up
Introduction
A comprehensive university data analysis project using the World Happiness Report dataset (156 countries, 9 variables) to investigate the structural, relational, and predictive dimensions of global happiness.
The analysis was structured across four areas:
Exploratory Data Analysis: Computed Pearson correlations to identify the strongest predictors of happiness. GDP per capita (r = 0.80), healthy life expectancy (r = 0.78), and social support (r = 0.75) emerged as the top drivers, while generosity showed minimal impact.
Relationship & Impact Analysis: Investigated whether rank–score relationships are linear or non-linear using LOWESS smoothing, revealing unequal sensitivity at ranking extremes. Showed that high freedom only translates to higher happiness when perceived corruption is low, uncovering a joint interaction effect. Identified that bottom-ranked countries score consistently below global averages across all structural dimensions.
Distribution & Variability Analysis: Confirmed happiness scores are approximately normally distributed (skewness ≈ 0.015). Demonstrated that top-ranked countries exhibit compressed score variance due to a ceiling effect and uniformly strong structural conditions, while low-ranked countries show higher volatility due to diverse patterns of structural deficits.
Predictive Modelling: Compared four models — Linear Regression, Random Forest, Linear Regression + PCA, and Random Forest + PCA — evaluating R² and RMSE across train/test splits. Used combinatorial search across all 3-variable subsets to identify the minimum feature set sufficient for strong prediction. Ran 30 repeated train–test splits to measure out-of-sample coefficient stability, finding that Generosity is the most unstable predictor across different samples.