Time Series Analysis and Forecasting Flashcards

What capability does a SARIMA model add over a standard ARIMA model?

Answer: Explicitly capturing repeating seasonal patterns

SARIMA (Seasonal ARIMA) extends ARIMA by incorporating additional seasonal AR, differencing, and MA terms to model patterns that repeat at fixed seasonal intervals.

Why is walk-forward (time series cross-validation) preferred over standard k-fold cross-validation for time series data?

Answer: It respects temporal ordering and prevents future data from leaking into the training set

Walk-forward validation trains on past data and evaluates on sequentially later windows, preserving time order and preventing data leakage that would inflate accuracy estimates.

Facebook Prophet is best described as which type of forecasting tool?

Answer: A decomposable additive model that handles trends, seasonality, and holiday effects automatically

Prophet is an additive model that decomposes time series into trend, multiple seasonality components, and user-specified holiday effects, designed to work well on business forecasting problems.

In time series analysis, what does 'cointegration' describe?

Answer: A long-run equilibrium relationship between two or more individually non-stationary series

Cointegration exists when two or more non-stationary series share a long-run equilibrium such that a linear combination of them is stationary, despite each wandering individually.

Why are Long Short-Term Memory (LSTM) networks preferred over standard RNNs for long time series sequences?

Answer: LSTMs use gating mechanisms to solve the vanishing gradient problem, enabling learning of long-range dependencies

LSTMs employ input, forget, and output gates that control information flow through the network, solving the vanishing gradient problem that prevents standard RNNs from learning distant dependencies.

What does Mean Absolute Percentage Error (MAPE) express about forecast accuracy?

Answer: Forecast errors expressed as a percentage of actual values, making accuracy scale-independent

MAPE expresses each error as a fraction of the actual value and averages these percentages, enabling meaningful accuracy comparisons across series with different scales or units.

Which is a well-known limitation of using MAPE as a time series forecast accuracy metric?

Answer: MAPE is undefined when actual values equal zero and penalizes over-forecasts more than under-forecasts

MAPE involves dividing by actual values, so it is undefined at zero; it also treats positive and negative errors asymmetrically, over-penalizing forecasts that exceed actual values.