Observations
To summarize, in this research I utilized an extensive data set of Bitcoin transactions to discover whether there are natural clusters with strong correlation to ransomware occurrences. I developed a script and used it for trying to segment the data set both for all 29 original labels and also for a binary partitioning. I used the K-means clustering algorithm in the attempts of finding a clear separation between suspicious and not suspicious transactions.
Although for both scenarios the plain Rand Index is close to 1, when evaluating the similarity adjusting for chance the scores drop below zero, which indicates low reliability. This leads me to conclude the combination of features, algorithm and consequently the generated model used do not represent a good method for classifying suspicious transactions in the Bitcoin Blockchain.
For future work I would consider comparing the results when using the address attribute, and also trying out different classification algorithms.