time series

Question 1

What is a time series?

Answer 1

A sequence of observations recorded at regular time intervals.

Question 2

What are the two main types of features in time series modeling?

Answer 2

Time-step features and lag features.

Question 3

What is a time-step feature?

Answer 3

A feature derived directly from the time index (e.g., time dummy).

Question 4

What is a lag feature?

Answer 4

A feature created by shifting the target series values backward in time.

Question 5

What is serial dependence?

Answer 5

When future values depend on past observations (modeled with lag features).

Question 6

What is time dependence?

Answer 6

When values depend on the time index (modeled with time-step features).

Question 7

What is trend in a time series?

Answer 7

A persistent long-term change in the mean of the series.

Question 8

How can you visualize trend?

Answer 8

Using a moving average plot with a window larger than seasonal periods.

Question 9

What is seasonality?

Answer 9

Regular, periodic changes in the mean of a series (e.g., daily, weekly, yearly).

Question 10

What are seasonal indicators?

Answer 10

Binary features representing seasonal periods (like one-hot encoded days of the week).

Question 11

What are Fourier features used for?

Answer 11

To model seasonality with many observations using sine and cosine curves.

Question 12

What is a periodogram?

Answer 12

A plot showing the strength of frequencies in a time series; helps choose Fourier pairs.

Question 13

What is a cycle in time series?

Answer 13

Patterns of growth/decay not tied to calendar time, often irregular.

Question 14

How can you visualize serial dependence?

Answer 14

Using a lag plot (current values vs. lagged values).

Question 15

What is autocorrelation?

Answer 15

Correlation of a time series with its lagged versions.

Question 16

What is partial autocorrelation?

Answer 16

Correlation of a time series with a lag, accounting for all shorter lags.

Question 17

What is a hybrid model in forecasting?

Answer 17

Combining two or more models to leverage their complementary strengths.

Question 18

What is a boosted hybrid?

Answer 18

Using one model to fit residuals of another (e.g., linear regression + XGBoost).

Question 19

What is a stacked hybrid?

Answer 19

Using predictions of one model as features for another model.

Question 20

What is forecast origin?

Answer 20

The last time point with available data used to make a forecast.

Question 21

What is forecast horizon?

Answer 21

The future time period for which forecasts are made.

Question 22

What is lead time (latency) in forecasting?

Answer 22

The time between forecast origin and forecast horizon.

Question 23

What is a multioutput forecasting strategy?

Answer 23

A single model that predicts multiple steps ahead simultaneously.

Question 24

What is the direct forecasting strategy?

Answer 24

Training separate models for each step in the forecast horizon.

Question 25

What is the recursive forecasting strategy?

Answer 25

Using a one-step model and feeding its forecasts back as inputs for next steps.

Question 26

What is the DirRec forecasting strategy?

Answer 26

Combining direct and recursive: separate models per step, using prior forecasts as features.

Question 27

What is a moving average?

Answer 27

The average of values within a sliding time window, used to smooth series.

Question 28

What is deterministic trend modeling?

Answer 28

Using polynomial functions of time (linear, quadratic, etc.) to model trend.

Question 29

What is the role of the DeterministicProcess function?

Answer 29

To create deterministic features like trend and seasonality without collinearity issues.

Question 30

How do you handle missing values when creating lag features?

Answer 30

Either fill them (e.g., with 0) or drop rows with missing values.

Question 31

What is a correlogram?

Answer 31

A plot of autocorrelation or partial autocorrelation against lags.

Question 32

Why can't tree-based models extrapolate trends?

Answer 32

They predict based on averages within training data ranges, not beyond them.

Question 33

How can you model multiple time series with XGBoost?

Answer 33

Convert data from wide to long format and add a categorical series identifier.

Question 34

What is the purpose of detrending in hybrid modeling?

Answer 34

To remove trend so another model (e.g., XGBoost) can focus on remaining patterns.

Question 35

What are leading indicators in forecasting?

Answer 35

Time series that change before the target series, providing early signals.

Question 36

How do you create lag features for multiple series?

Answer 36

Shift each series individually and align them into a feature matrix.

Question 37

What is the difference between seasonality and cycles?

Answer 37

Seasonality is calendar-based and regular; cycles are irregular and not tied to dates.

Question 38

What is the additive model of time series?

Answer 38

Series = Trend + Seasonality + Cycles + Error.

Question 39

What is the residual of a model?

Answer 39

The difference between actual values and model predictions.

Question 40

How do you prepare data for multistep forecasting?

Answer 40

Create multiple target columns (one per step) and align with lagged features.

Question 41

What is a common use of Fourier features?

Answer 41

To model annual seasonality with daily data using few features.

Question 42

What is the advantage of using linear regression for trend?

Answer 42

It can extrapolate trends beyond training data given appropriate time features.

Question 43

What is a lag plot used for?

Answer 43

To visualize the relationship between a series and its lagged version.

Question 44

What does a seasonal plot show?

Answer 44

Segments of a time series aligned by a seasonal period (e.g., weeks, months).

Question 45

What is the smoothing parameter in moving averages?

Answer 45

The window size, which determines how many periods are averaged.

Question 46

How do you choose the number of Fourier pairs?

Answer 46

Using the periodogram to identify dominant frequencies in the series.

Question 47

What is the main limitation of using only lag features?

Answer 47

You can only forecast up to the number of steps for which lag values are available.

Question 48

What is the benefit of hybrid models?

Answer 48

They combine trend extrapolation (linear models) with pattern learning (tree/neural models).