I am relativly new to the field of data mining and want to make a anomaly detection on transactional retail data. I want to use a simple anomaly detection (kmeans at the moment) for finding suspicious transactions with operators.
For example I want to use the following attributes:
- Counter of his transactions
- Counter of his discount transactions
Now I ask myself:
- Is it possible to detect outliers/anomalies with this kind of comparison (Sales / Discount Transactions)
- Is the K-Means Anomaly Detection the right algorithm for this kind of problem?
- How does the algorithm deal with relationships?
- Does it make sense to use this attributes?
- Do you think a classification task would work out better? Transactional data is very different.
- Do you have other suggestions regarding attributes, dimensions, algorithms?