I have been lookig to the following notebook: time series clustering
where the writer says that the dataset is affected by the “Curse of Dimensionality”, so applying TimeSeriesKMeans offered by tslearn is not correct. Instead, PCA is applied on the column “values” of the timeseries to get just 2 components out of the time-series values and then a normal k-means function is used to divde into clusters.

I have been looking online and seems that the Curse happens only when the dataset have many features and few data. The dataset in the article has only 2 feature (timestamp, value) and many data, so in my opinion there is no “Curse of dimensionality”, hence doing the PCA on all the “value” column is useless.

Am I missing something important?

On the other hand, if the article is right, every time I want to study a time series I may just apply PCA on all the values and then apply the classic k-means, instead of the TimeSeriesKMeans function offered by tslearn