Analyzing Patterns / Trend

mustafatd

I am completely new to Rapid Miner. I have a problem statement on hand but don't know where to start.

Problem statement:
I have few banking transactions as a data set and i need to profile the transaction pattern of each of the accounts. Then every new transaction will be checked against the account's profile and if there is any deviation from the original profile, it should throw up as an exception.

Please let me know which algorithm is best suited for the above situation and how to go about it in Rapid Miner.

Thanks

Mustafa

fischer

Hi,

I am afraid, this is too general a question to answer it in a single post. If you just want to measure the similarity to other transactions as you are writing, filter by account, compute the similarity to all using "Data to Similarity", and aggregate, computing, e.g., the maximum distance or the largest distance. Alternatively, filter by account and use an outlier detection algorithm to see whether your new transaction differs from the others. However, none of these may be what you exactly want, but this can be answered only by yourself and it is probably best to first get a bit familiar with RapidMiner and data mining.

Cheers,
Simon