KRASI1234/ML-and-Food-Production

Machine Learning Forecasting of Agricultural Production and Food Distribution in Ghana # Machine Learning Forecasting of Crop Production and Food Prices in Ghana ## Overview Statistical and machine-learning models forecasting crop production and staple-food prices in Ghana. Compares ARIMA, simple benchmarks (Naïve, Drift), and machine-learning models (Ridge, Random Forest), selecting the best-performing approach per crop based on out-of-sample validation rather than in-sample fit. Four staple crops: Maize, Cassava, Rice, Yams. ## Objectives - Forecast production for selected staple crops in Ghana. - Forecast staple-food prices using historical and socioeconomic data. - Compare statistical and machine-learning forecasting models. - Provide data-driven recommendations for policymakers. ## Tools R, Python/Jupyter, VS Code, Git and GitHub. ## Project Status ### Phase 1: Project Setup Repo and folder structure created; Git and Jupyter notebooks initialized. ### Phase 2: Data Collection Sourced FAOSTAT crop-production data, checked availability and coverage, selected the four study crops. ### Phase 3: Data Cleaning Cleaned raw data, standardized crop names and formats, reshaped into analytical structure, saved cleaned dataset. ### Phase 4: Exploratory Data Analysis Examined structure and descriptive statistics; production, harvested area, and yield across crops; trends, relationships, correlations. ### Phase 5.1: Crop-Specific Dataset Preparation Separate datasets built for Maize, Cassava, Rice, Yams. Chronologically sorted, checked for year continuity and missing observations, saved as forecasting datasets. ### Phase 5.2: Time-Series Diagnostics Constructed annual series per crop. ADF and KPSS tests for stationarity, Ljung-Box for serial dependence, ACF/PACF to inform ARIMA specification. Reassessed after first differencing; log-transform plus differencing checked for scale/growth effects. Diagnostics run independently in R and Python as a cross-check, with Python as the main modelling environment going forward. No seasonal decompositio …