Machine Learning Analysis of the World Happiness Report: Predictions for 168 Countries Through 2030

This project is based on the dataset of the United Nations' World Happiness Report, using machine learning technologies such as PyTorch and scikit-learn to conduct in-depth analysis of multi-dimensional happiness indicators of 168 countries, and builds predictive models to forecast global happiness trends up to 2030. Core questions explored: Which factors best predict a country's happiness level? How will global happiness trends evolve over the next decade? Are there differences in happiness drivers across regions?

Happiness from philosophical proposition to data science: In the past decade, big data and machine learning technologies have promoted quantitative research on happiness. The United Nations SDSN has released the World Happiness Report since 2012, covering multi-dimensional indicators such as GDP per capita, social support, and healthy life expectancy. The dataset for this project covers data from 168 countries from 2005 to 2024, with core features including life evaluation (Cantril Ladder), log of GDP per capita, social support, healthy life expectancy, freedom, generosity, perception of corruption, positive/negative emotions, etc.

Prediction tasks: Regression task (predicting happiness scores) + time series prediction (trends up to 2030). Model selection: Baseline models (linear/ridge regression), ensemble models (random forest, XGBoost/LightGBM), neural networks (PyTorch-MLP), time series models (ARIMA, Prophet, LSTM). Feature engineering: Lag features (happiness score/GDP lag terms), interaction features (GDP and social support, etc.), regional encoding, time features (year/decade interval).

Exploratory Analysis: Nordic countries have long been leading (high social support, low corruption); regional differences are significant (North America and Western Europe are at the top, while Sub-Saharan Africa is lower); the global average happiness is slowly rising but growth is uneven. Correlation: GDP and happiness have a non-linear relationship (diminishing marginal effect); social support has the strongest correlation with happiness; perception of corruption is negatively correlated. Model Results: Gradient boosting trees have high R²; feature importance ranking: social support > GDP per capita > healthy life expectancy > freedom > perception of corruption > generosity; regional differences: developed countries pay more attention to soft factors, while economic factors are more important in developing countries.

2030 Predictions: The global average happiness will continue to rise; regional gaps may narrow (developing countries in Asia are growing rapidly); the importance of social support will be highlighted; climate change will have a negative impact. Core Findings: Happiness is jointly determined by multiple dimensions such as economic development, social support, health, personal freedom, and institutional quality; single economic growth is not sufficient to improve happiness.

Policy Implications: Go beyond the GDP-oriented development concept, focus on social support networks, medical systems, and institutional quality; invest in social infrastructure (community building, mental health services); strengthen anti-corruption efforts. Personal Implications: Maintain family and friend relationships; participate in volunteer/charity activities; attach importance to health management.

Current Limitations: Data relies on self-reports (cultural bias); observational data makes it difficult to determine causality; long-term prediction uncertainty is high. Future Directions: Integrate city/individual-level data; adopt causal inference methods; build real-time happiness monitoring systems; cross-disciplinary integration of psychology and sociology theories.