Forecasting cards

Question 1

ways to decompose seasonality from data

Answer 1

classical seasonal decomposition, X11 decomposition, SEATS decomposition

Question 2

ARIMA

Answer 2

Autoregressive, integrated moving average

Question 3

Establish stationarity

Answer 3

Rolling statistics (rolling mean should be similar to zero)
ADF test

Question 4

Autoregressive model

Answer 4

a model that assumes output is based on its own previous values and a stochastic term

Question 5

Augmented Dickey–Fuller test

Answer 5

Test to determine if unit root. The more negative, the less likely a unit root.
If there is a unit root present in the time series (null hypothesis), it implies that the time series is non-stationary.
If the null hypothesis is not rejected, the time series is non-stationary. If the null hypothesis is rejected, it’s stationary.

Question 6

KPSS test

Answer 6

Test for stationarity of a time series. Null hypothesis is that time series is stationary.

Question 7

K-means clustering

Answer 7

a process that assigns each data point to its nearest center point and then recomputes the mean of each group.
Elbow method determines the best number of clusters, by mapping out average distance to each cluster center

Question 8

DBSCAN

Answer 8

Creates clusters by iteratively going over each point and if there is enough other points within a certain distance, a cluster is started and it continues to take other points that satisfy the criteria within the cluster. Points that do not reach the threshold are classified as noise