About graph analytics

The graph analytics component allow users to load data, run feature engineering, and run machine learning modes to compute the network score. The network score attributes to the risk factor associated with the account based on the transaction patterns and closeness to the accounts, which are flagged for a suspicious activity report (SAR).

The component loads party data, entity resolution output (to include non-obvious relationship between parties in the system), accounts, and related transactions data. The transaction data can be loaded as raw data and also aggregated (for example, by weekly, by monthly, and so on)

The graph is comprised of party-relations as well as transactional data. It allows the user to run various algorithms (Degree, Label Propagation, Ego Network, Page Rank, Temporal Cycle Detection, and Risk by Association) to generate features that can help in associating the network score to a given account or party.

The graph component applies a machine learning method to learn the pattern and predict the suspicious activity in any given account. It uses an XGBoost model to classify suspicious activity versus normal activity in the account. The output of the machine learning model is translated to network score and also stored in the graph.

The visualization component allows users in an Admin or Supervisor role to look at all the high risk accounts / parties and be able to explore the associated network with the account to understand the suspicious activity corresponding to the account. It also displays associated parties to the account and its association. This component lets the user filter based on time range, transaction type, and number of hops. It also allows the user to expand on-demand and explore the transaction network.