Clusters multiple time series data into groups.
Input data should be a time series data with category. Each row should represent one observation with date/time. It may have multiple rows for a date/time, in which case the rows are internally aggregated into one row for the date/time. It should have the following columns.
Group - A categorical (character or factor) column. The categories specified here are clustered into groups.
Date/Time - A Date or POSIXct column to indicate when the observations took place.
Value (Optional) - A column that stores observed values. Values for multiple rows for one date/time for a category are internally aggregated into one value by the specified aggregation function to form a time series for the category to be clustered. If not specified, the number of rows for each date/time is used as the time series to cluster.
Other Columns to Keep (Optional) - Other columns for values to keep in the output data. Values for multiple rows for one date/time for a category are internally aggregated into one value by the specified aggregation function, to be put together in the output.
Number of Clusters - The number of clusters to group the time series data into.
Cluster Center Method - Method to calculate cluster center time series (centroid) for each iteration.
DTW Barycenter Averaging
Soft DTW Centroids
Partition around Medoids
Distance Method - Method to calculate distance between the cluster center time series (centroid) and each time series for each iteration.
DTW with L2 Norm
DTW Guided by Lemire's Lower Bound - 10% of the length of the data is set for the window size.
Keogh's Lower Bound for DTW - 10% of the length of the data is set for the window size.
Lemire's Lower Bound for DTW - 10% of the length of the data is set for the window size.
Global Alignment Kernels
Random Seed - Random seed set before the clustering, so that the results are constant when the same calculations are repeated.
NA Fill Type - How to fill NAs that appear between the first and last non-NA value in a time series.
Fill with Previous Non-NA Value
Fill with 0
NA Fill Type - Beginning - How to fill NAs that appear before the first non-NA value in a time series.
Fill with 0
Fill with First Non-NA Value
NA Fill Type - Ending - How to fill NAs that appear after the last non-NA value in a time series.
Fill with 0
Fill with Last Non-NA Value
Remove Groups with NAs
When NA Ratio Is Greater Than - If the time series data for a category has more NAs than this ratio, the category is removed from the data before the clustering is performed.
Normalize Value - Whether to normalize the aggregated values or not.
Time Series Clustering Step uses the dtwclust R Package under the hood.
For details about
dtwclust usage in Exploratory R Package, please refer to the github repository