Why use XGBoost for time series forecasting?

In time series forecasting, selecting the right model is essential to capturing the underlying patterns and making accurate forecasts. Statistical time series models, such as Autoregressive Integrated Moving Average (ARIMA) and Seasonal ARIMA (SARIMA), and exponential smoothing methods are often effective for stable data with consistent trends and seasonality. However, they tend to perform poorly when the data exhibits frequent spikes or sudden changes. These models generally assume linear relationships and are designed to work best with stationary data—data without large fluctuations in mean or variance. When faced with high volatility or abrupt shifts, such models may struggle to capture the underlying structure accurately, leading to less reliable forecasts (Box, Jenkins, and Reinsel, 2015; Brockwell and Davis, 2016). In contrast, XGBoost’s flexibility with non-linear relationships and robustness to outliers make it better...