Built an end-to-end forecasting pipeline in R to predict May–June 2026 monthly sales for a Tunisian retail store. Starting from raw ERP data (Jan 2022–Apr 2026), I handled preprocessing, stationarity testing (ADF/KPSS), and implemented four models — SARIMA, ETS, Prophet, and XGBoost — evaluated on a held-out test set. Prophet achieved the best MAPE (8.53%), while XGBoost captured short-term nonlinear patterns most closely. The project highlighted the tradeoffs between statistical interpretability and ML flexibility on a short, event-driven dataset.