Table 1.

Benefits and limitations of selected forecasting models.

Models	Benefits	Limitations
SARIMA/ SARIMAX	Solid mathematical and statistical theory. Time-varying trends/seasonal patterns. Relatively few parameters. Handles exogenous variables.	Difficulty tuning the model parameters. Usually computationally expensive. Prone to overfitting.
FP	Supports seasonality with multiple periods. Robust to missing data. Does not require data interpolation. Handles outliers. Handles exogenous variables.	Does not consider multiplicative models. Strict formatting requirement Restricted to Gaussian noise distribution. Does not take autocorrelation into account. Does not assume a stochastic trend.
HW	Works best for data with trends and with seasonality that increases over time. The results are interpretable. Very easy to implement.	The presence of outliers distorts the results. Not expanded to multivariable approach. Accounts for only a single seasonal pattern.
LSTM	Learns information for an extended period. Mitigates the vanishing gradient problem. No specific assumptions. Handles exogenous variables.	Computationally time-consuming. Sensitive to random weight initializations. Prone to overfitting.

SARIMA: seasonal autoregressive integrated moving average; FP: Facebook Prophet; HW: Holt-Winters; LSTM: Long Short-Term Memory; and SARIMAX: seasonal autoregressive integrated moving average exogenous.