lavsz/Captone_Project-Africa_Food_Security_and_Infrastructure

A project for diving into sustainability in Africa. It composes: building sustainability index on Admin1-level; studying top contributors of Food Security Classification # Project: Africa Development Potentials and Food Security by Shenghao (Lavender) Zhang The year of Covid-19 strengthens more about Sustainability Development worldwide and disease spreading prevention made people start to think more about food security. ## Overview This project cooks a two-course meal: 1. Studying geophysical and social factors that contributes to Food Security concerns in selected region 2. Looking at infrastructure development potential **The first course**: aims to built a classification model that can identifies areas with higher food security concerns. **The second course**: aims to look at different geophysical, accessibility, facilities, and social factors to propose a new evaluation method for development potentials in the infrastructure sector. ## First Course: Food Security Machine Learning Model **How's the distribution of different food security class?** **On Average, how are different countries throughout the year looks like?** Note: Madagascar was not evaluated from 2010 to 2015. **Are there many social conflict events happening?** Protests went by a lot in 2020. **Machine Learning Models** To classify the severity level of food security in Africa, time-specific geophysical data, nighttime light, and social conflict records were added to the model. After the first evaluation, it was noticed that imbalanced classes are really affecting the modeling results. The goal is the model is to be able to alert regions that would have high possibility in severe food crisis. To better separate very severe vs less severe conditions, Class 1 (minimal) and Class 2 (stressed) are being grouped into Class 2 (less severe); Class 3 (crisis), Class 4 (emergency), and Class 5 (famine) are group into Class 3 (very severe). Two winning model algorithms are: XGBoost - 81.04% Random Forest Classifier = 80.8% XGBoost is slightly stronger with also less discrepancies between training and testing sets. **What are the top contributors?** …