RohinM24/Predicting-Year-to-Year-Changes-in-Suicide-Rates-in-South-Africa-

I built a panel-trained model to predict year-to-year changes in suicide rates for South Africa. Structural socio-economic indicators explain long-run trends but provide limited predictive power for short-term fluctuations, which are dominated by unobserved shocks. Please download the zip files and open in on Google Colab, or Jupyter Notebook Predicting Year-to-Year Changes in Suicide Rates in South Africa A Panel-Based Socioeconomic Modeling Approach 📌 Project Overview This project develops a data-driven, statistically grounded machine learning model to predict year-to-year changes in suicide rates in South Africa, using socioeconomic signals learned from a multi-country panel dataset. Rather than predicting absolute suicide levels, the model focuses on annual changes, which: Reduces non-stationarity Avoids spurious trend learning Aligns with best practices in epidemiological and economic time-series modeling The project emphasizes interpretability, robustness, and uncertainty awareness, making it suitable for research, policy analysis, and academic work. 🎯 Objectives Predict annual changes in suicide rates for South Africa Learn structural relationships from comparable countries using panel data Avoid temporal leakage using time-aware validation Quantify uncertainty in predictions Produce an interpretable and defensible baseline model 🧠 Methodological Summary Target Variable SuicideRate_diff Year-over-year change in suicide mortality rate (per 100,000 population) Predictors (3-Year Rolling Averages) Alcohol consumption per capita Intentional homicide rate GDP per capita (current US$) Rolling averages are used to capture structural trends while reducing short-term noise. 📊 Data Sources All data comes from authoritative international sources: Suicide mortality rate – UN SDG / WHO Alcohol consumption – WHO / UN SDG Homicide rate – UN SDG GDP per capita – World Bank (NY.GDP.PCAP.CD) Countries were selected based on: Data availability (2000–2024) Reporting consistency Socioeconomic comparability South Africa is explicitly held out as the prediction target. 🧩 Modeling Approach Model type: Ridge Regression (L2-regularized linear model) Training strategy: Panel learning on multiple countries Sou …