Asma-Hassan9/somalia-food-price-forecasting

Machine learning project forecasting food price volatility in Somalia and generating proxy early-warning alerts for vulnerable regions and commodities. # Machine Learning Forecasting of Food Price Volatility in Somalia An applied machine-learning project that forecasts food commodity prices and converts predicted price levels into a proxy early-warning system for Somalia. ## Project overview Somalia's food markets are affected by drought, floods, supply-chain disruptions, exchange-rate movements, and fragmented market reporting. This project evaluates whether machine-learning models can forecast food commodity price volatility using historical market prices and exchange-rate data, and whether those forecasts can be translated into practical risk signals for policymakers and humanitarian organizations. The project compares Random Forest, XGBoost, LightGBM, LSTM, and a hybrid LSTM-XGBoost model. The strongest model is then used to classify predicted prices into four alert levels: Normal, Warning, Alert, and Crisis. ## Objectives - Prepare and integrate Somalia food-price data with exchange-rate data. - Engineer time-series, volatility, commodity, market, and regional features. - Compare tree-based, gradient-boosting, deep-learning, and hybrid models. - Identify the most important drivers of predicted food prices. - Create a price-threshold proxy early-warning mechanism. - Present regional and commodity risks through an interactive dashboard. ## Data sources - World Food Programme - Somalia Food Prices - World Bank Microdata Library The modelling dataset combines historical food-price observations with Somalia's unofficial exchange rate. Raw source data is not redistributed in this repository; users should retrieve it from the original providers and follow their licensing terms. ## Methodology The project follows the CRISP-DM framework: 1. Business and data understanding 2. Data cleaning and integration 3. Missing-value treatment and outlier analysis 4. Log transformation, encoding, scaling, and feature engineering 5. Feature selection using correlation, multicollinearity checks, F-tests, and mutual inform …