averagenomad/Africa-Time-Series-Forecasting

# **Time Series Analysis of Civil Conflict: Africa (ARIMA Modeling)** ## **Overview** This repository documents an exploratory project in applying **time series analysis** to civil conflict event data in Africa. The goal was to investigate whether historical patterns in conflict data alone could be used to predict future events, using **Box-Jenkins ARIMA modeling**. The analysis focused on four African countries experiencing civil conflict, with a **full end-to-end modeling process demonstrated for Ethiopia and Somalia**. ### **Key Insights** - Out-of-sample forecasting performance varied by country: - **Somalia:** ARIMA(3,1,0) performed well with low RMSE values. - **Ethiopia:** ARIMA(1,1,1) was the best fit but less effective in forecasting. - Results demonstrate that **simple ARIMA models can deliver strong forecasting accuracy** even when relying solely on past events, though predictive performance is highly country-dependent. --- ## **Repository Structure** ``` ├── data/ │ ├── Ethiopia.csv │ └── Somalia.csv │ ├── notebooks/ │ └── africa-time-series.ipynb │ └── output/ └── IST 341_Final Presentation_Zhamilia Klycheva(Jama).pdf ``` - **data/** – Monthly time series event counts for Ethiopia and Somalia (subset of ACLED conflict event data). - **notebooks/** – Jupyter Notebook implementing data cleaning, exploratory data analysis (EDA), and ARIMA modeling. - **output/** – Final project presentation summarizing results and methodology. --- ## **Data & Methodology** - **Data source:** Armed Conflict Location Event Data Project (ACLED). - Each country’s data was grouped at a **monthly frequency**, removing non-violent events. - The **Box-Jenkins ARIMA approach** was applied: 1. Model identification (ACF/PACF plots, differencing) 2. Parameter estimation and evaluation using AIC/BIC 3. Diagnostic checks and rolling forecast evaluation --- ## **Future Extensions** - Ex …