Time Series Analysis Flashcards
7 cards from real Data Science practice questions. Tap to flip, then mark Knew It or Still Learning — missed cards come back until you master them.
Read the first 7 Time Series Analysis flashcards as text
How does SARIMA differ from a standard ARIMA model?
Answer: SARIMA adds seasonal AR, differencing, and MA terms to capture seasonal patterns
SARIMA extends ARIMA by adding seasonal autoregressive, integrated, and moving average terms (SARIMA(p,d,q)(P,D,Q)m) to model periodic seasonal patterns alongside non-seasonal structure.
Why is standard k-fold cross-validation inappropriate for time series forecasting evaluation?
Answer: It violates temporal ordering by using future data to train models predicting the past
Standard k-fold randomly shuffles observations, causing data leakage where future values influence training — this artificially inflates performance and is invalid for time-dependent data.
What does the Granger causality test assess in time series analysis?
Answer: Whether past values of one series are useful for forecasting another series
Granger causality tests whether including lagged values of series X improves forecast accuracy of series Y beyond using Y's own history alone, indicating predictive (not causal) relationships.
Long Short-Term Memory (LSTM) networks are well-suited for time series forecasting because:
Answer: They learn long-range dependencies and model complex non-linear patterns
LSTMs use gating mechanisms (input, forget, output gates) to selectively retain or discard information across long sequences, capturing complex temporal dependencies that linear statistical models cannot represent.
Cointegration between two non-stationary time series implies:
Answer: A stable long-run equilibrium relationship exists between the two series
Cointegration means that even though each series individually is non-stationary, a linear combination of them is stationary — indicating a persistent, stable long-run equilibrium relationship.
The Facebook Prophet model for time series forecasting is based on:
Answer: An additive regression model combining trend, seasonality, and holiday effects
Prophet uses an additive decomposition framework, modeling time series as the sum of a piecewise trend, Fourier-based seasonal components, and user-specified holiday effects.
What is the 'walk-forward' (expanding window) validation strategy in time series?
Answer: Iteratively training on growing historical data and predicting the immediately following period
Walk-forward validation trains on all data up to time t, predicts period t+1, then advances — simulating real forecasting conditions while strictly respecting temporal ordering to prevent data leakage.