The machine learning model 130 may also be used to predict whether a risk of fraud associated with the connection request is low or high, and the machine learning model 130 may output a trust score classification that corresponds to the predicted risk. The scoring agent 110 may return the trust score classification output by the machine learning model 130 to the communication service 108, and the communication service may use the trust score classification to allow or deny a connection request associated with the customer account.

In the example described above and illustrated in FIG. 1, the scoring agent 110 generates a trust score used to determine whether to allow a connection request associated with a customer account. In another example, as illustrated in FIG. 2, the communication service 108 may be configured to generate a trust score in response to receiving a connection request. For example, the communication service 108 may be configured to generate a trust score using a rule set 126, call pattern 128, and/or machine learning model 130 as described above with respect to the scoring agent 110. As an illustration, the communication service 108 may receive a connection request from a caller, generate a trust score using any of the techniques described above, and make a determination whether to allow the connection request based on the trust score.