I have 3 years of continued data of time series in 5 min time interval - 50k rows. I have extracted a list of 1561 samples from the data, defined by exact mathematical conditions. Each record has just date and time of day.
One record in the list of samples - is one sample that contains “2x27 time steps of the past values” in the 5 minute time interval. The samples do not exist yet.
The samples will be clustered to 16 clusters. The final result that is needed after clustering is - After clustering will be needed -
a) to know the position of each sample in the particular cluster - in the original list of samples;
b) to know about the first 3 clusters where are most samples located after clustering.
Here are my questions:
1.) What will be the best clustering method for time series (like SOM, hierarchical..) and what will be the best an Open program & language for it - like Somoclu, Orange (Silhouette), Spark,.. to get the required result after clustering?
2.) Does it need to create the database of the samples from the historical data for the clustering and if Yes, what free database solution will be the best? .. or is there any chance that a Python or R expert can code “an add-on like” for the instant connection between clustering program and historical data followed by the list of samples (for the process of clustering and final result too)?