I am currently trying to expand upon some results from the paper "Time Warping Clustering for the Forecast and Analysis of COVID-19" by Qixuan Jin out of Cal Tech, because when it was published they only had 4 months of data. The paper is essentially looking for clusters of covid cases at the FIPS code level. In the methods step the last preprocessing step before it goes into how it clustered talks about "tier splitting the data". I have searched EVERY WHERE looking for what this means and I can not find anything other than when working with customer data you can used and RFM score and tier split based on that. However, since covid cases doesn't related to customer data, I don't think this is what it is talking about. My data frame has 3219 columns with each column a different FIPS code; it has 470 rows each row a different day with the number of covid cases per day. I went a head and totaled the number of cases over the 1 and 104 days for each fips code. Here is what it says about tier splitting :

"Tier Splitting: Magnitude can still be used to guide clustering at a higher level. Counties are split into tiers based on their cumulative case counts. Clustering is then performed within each tier. We propose two tier-splitting methods: splitting into four quartiles and splitting based on the custom thresholds of 10, 100, and 1000 cumulative cases."

Am I understanding this correctly that for one method they went ahead got the quartile values for the cumulative cases and then split up the data frame based on the various quartiles and then clustered the various quartile groups. Then for the custom split the split based on 0-10 total cases, 11-100 total cases, 111-1000 total cases, and 1001+ total cases? Or am I misundertanding?

Thank you


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.