This table estimates the relative individual importance of each feature as a substitute for its importance in the overall clustering model, assigning a score to each feature.

Cluster Model

The name of the cluster definition.

Time Stamp

The date and time at which the cluster definition was run by this action.

Feature

The clustering feature that was extracted during execution.

Importance Score

The relative individual importance of the feature as a substitute for its importance in the overall clustering model.

Selected

Indicates if the feature was chosen as a part of the final clustering model.

Model ID

The internal ID of the model.

This table displays a score for overall quality of each cluster run by this action, indicating how well the cluster represents the time series it contains.

Cluster Model

The name of the cluster definition.

Time Stamp

The date and time at which the cluster definition was run by this action.

Cluster Quality Score

The overall score for the cluster.

Model ID

The internal ID of the model.

The Demand Guru Clustering action outputs two cluster feature tables with the same columns:

• Scaled - The data is normalized between -1 and 1.
• Unscaled - The data is not normalized.

Cluster Model

The name of the cluster definition.

Time Stamp

The date and time at which the cluster was run by this action.

Time Series

The user-specified name of the time series; for example, SKU identifiers, customer identifiers, or any of the "Group by" tags used in the data being clustered.

Cluster

The cluster to which the time series belongs after execution of the cluster model.

Representative Time Series

Indicates if the time series is representative of the cluster in which it is classified.

Seasonality

A coefficient value based on the periodicity of the most prominent seasonal period in a time series. Low values indicate low frequency or long period seasons, while high values indicate high frequency or short periodic time series. An absence of seasonality in the data is assumed to indicate extremely high frequency (noisy data) and has a high value associated with it.

Trend

An index of the strength of a trend. High positive values indicate strong upward trend, high negative values indicate a strong downward trend, and values close to zero indicate that the trend is flat.

Mean

Mean of the demand values in a time series.

Variance

Variance of the demand values in a time series.

Auto Correlation

Represents the extent of dependence on past demand values. While calculating this score, 10 lags are considered. A higher dependence of the time series on its past 10 values results in a higher auto correlation score.

Lumpiness

The variability of variance of each period in a time series. Conceptually, divide the time series into multiple sections, calculate the variance of each section, and then calculate the variance of these variance values. Low values indicate that the variance of the time series does not change much across its different sections, while high values indicate the variance is changing a lot across different sections of a time series.

Level Shift

The maximum absolute value of mean values of slices of a time series, when a rolling window (of size =1) is used. Seasonal time series are divided into multiple slices, with the slice length equal to its most prominent period. For non-seasonal time series, this slice length is equal to a fixed constant.

The first slice is rolled by the window size (= 1 here), and the maximum absolute value of the mean of the rolled slice gives the value of the level shift.

Intuitively, this value represents the maximum “level” in a section of a time series.

Variance Change

The maximum absolute value of variance values of slices of a time series, when a rolling window (of size =1) is used. Seasonal time series are divided into multiple slices, with the slice length equal to its most prominent period. For non-seasonal time series, this slice length is equal to a fixed constant.

The first slice is rolled by the window size (= 1 here), and the maximum absolute value of the variance of the rolled slice gives the value of the variance change.

Intuitively, this value represents the maximum variance in a section of a time series.

Crossing Points

The number of times a time series crosses the midpoint of its range. Range is nothing but (Maximum value – Minimum value).

Linearity

Strength of the linearity component of a time series, with high values indicating strong linearity, and positive values indicating an upward trending time series.

Curvature

The strength of a trend's curvature component in a time series. Positive values indicate convex shaped time series, while negative values indicate concave shaped time series.

Peak

Strength of the highest point on the seasonal component of a time series.

Trough

Value of the lowest point on the seasonal component of a time series.

Entropy

A measure of the forecastability of a time series. It reveals the degree of difficulty associated with forecasting a specific time series, based on only the demand values. Low values indicate that the time series is relatively easy to forecast, and higher values indicate increased difficulty.

Spikiness

The strength of spikes of residuals in a time series. The seasonality and trend components are first removed, and then the value is calculated over the residual component.

Flat Spots

The length of flat spots in a time series. To arrive at this value, a time series is broken down into multiple discrete levels. Then, an analysis is made to determine the number periods for which the time series maintains the same level. The overall length of these periods provides the value for this feature. A lower value means that the time series is changing its discretized levels more often, and a higher value means that the time series is changing its levels less often.

Model ID

The internal ID of the cluster model.