HMMDecoder Example 4: Bank Customer Churn - Teradata Vantage

This example uses HMMDecoder to find the propensity of customer churn, given the actions or transactions of a bank customer.

Input

In the following three tables, T indicates that the customer is likely to churn and F indicates that the customer is unlikely to churn.

In the observation table, the column action has transactions, the column id has customer identifiers, and the column path_id shows transaction order.

SQL Call

SELECT * FROM HMMDecoder (
  ON churn_initial AS InitStateProb PARTITION BY model
  ON churn_state_transition as TransProb PARTITION BY model
  ON churn_emission AS EmissionProb PARTITION BY model
  ON churn_data AS observation PARTITION BY model
  ORDER BY model,id, path_id ASC
  InitStateModelColumn ('model')
  InitStateColumn ('tag')
  InitStateProbColumn ('probability')
  TransAttributeColumn ('model')
  TransFromStateColumn ('from_tag')
  TransToStateColumn ('to_tag')
  TransProbColumn ('probability')
  EmitModelColumn ('model')
  EmitStateColumn ('state')
  EmitObsColumn ('observed')
  EmitProbColumn ('probability')
  ModelColumn ('model')
  SeqColumn ('id')
  ObsColumn ('action')
  Accumulate ('path_id')
) ORDER BY 1, 2, 5, 3;

Output