Machine-learning based delay prediction
Aspects of the subject technology include obtaining a transfer request event associated with a transaction, the transfer request event indicating a request day of week and time of day, and, when the request time of day is past a pre-determined cutoff time, adjusting the request day of week to be a subsequent day. Aspects also include determining, using the transfer request event and based on a machine learning model trained on historical transaction data, a predicted transfer delay for the transaction, obtaining a set of relevant non-transfer days based on a comparison between the historical transaction data and a set of past non-transfer days, and, when one or more non-transfer days from the set of relevant non-transfer days occur within a time period from the request day of week and over the predicted transfer delay, adjusting the predicted transfer delay based on the one or more non-transfer days.
1 . A method comprising:
obtaining, by a first device, details of a network event, the details for the network event indicating a day of week associated with the network event and a time of day associated with the network event;
in response to determining that the time of day associated with the network event is past a pre-determined time of day, adjusting, by the first device, the day of week to be a subsequent day of week;
determining, by the first device, using the details of the network event and based on a machine learning model trained on historical network event data, a predicted verification delay for verifying the network event by a recipient of the network event, wherein training the machine learning model comprises: preprocessing the historical network event data including non-numeric categorical data and continuous data to encode the non-numeric categorical data into encoded data and to convert the continuous data into converted numerical data, and wherein the machine learning model is trained based on the encoded data and the converted numerical data;
obtaining, by the first device, a set of relevant non-verification days for the network event based at least in part on a comparison between the historical network event data and a set of past non-verification days, wherein the set of relevant non-verification days correspond to days on which verifying the network event by the recipient cannot occur;
in response to determining that one or more non-verification days from the set of relevant non-verification days occur within a time period from the day of week and over the predicted verification delay, by the first device, increasing the predicted verification delay based on the one or more non-verification days;
providing, by the first device, the predicted verification delay to one or more downstream services for performing a planned operation based on an estimated verification delay, the predicted verification delay corresponding to the estimated verification delay, wherein the downstream services are configured to perform processing resource planning based on the estimated verification delay, thereby resulting in a reduction of processing resource utilization upon verification of the network event; and
when an actual verification delay extends beyond the predicted verification delay, flagging, by the first device, the network event and triggering an alert for intervention.
2 . The method of claim 1 , further comprising, prior to determining the predicted verification delay:
determining a baseline verification delay for the network event based on a segment of the historical network event data; and
in response to the predicted verification delay being greater than the baseline verification delay, adjusting the day of week of the network event to a next day that is not a non-verification day.
3 . The method of claim 1 , wherein increasing the predicted verification delay comprises:
in response to the one or more non-verification days coinciding with the predicted verification delay, increasing the predicted verification delay by one day.
4 . The method of claim 1 , wherein increasing the predicted verification delay based on the one or more non-verification days comprises extending the predicted verification delay by one day for each of the one or more non-verification days.
5 . The method of claim 1 , further comprising updating one or more attributes of the network event based on the predicted verification delay.
6 . The method of claim 1 , further comprising:
receiving, from a remote server, reconciliation data associated with the network event;
determining a difference between the predicted verification delay and the actual verification delay indicated by the reconciliation data; and
re-training the machine learning model based on the difference.
7 . The method of claim 1 , wherein obtaining the set of relevant non-verification days comprises filtering the set of past non-verification days by at least a country associated with the network event.
8 . The method of claim 1 , wherein the set of past non-verification days includes past holidays.
9 . The method of claim 1 , wherein the one or more downstream services comprise at least one of a safeguarding service, a resource management service, an intraday alerting service, or a foreign resource service.
10 . The method of claim 1 , wherein the historical network event data includes a subset of available historical network event data, wherein the historical network event data is sampled from the available historical network event data in a stratified manner to ensure the historical network event data is representative of the available historical network event data.
11 . An electronic device comprising:
a memory; and
a processor configured to:
obtain details of a network event, the details for the network event indicating a day of week associated with the network event and a time of day associated with the network event;
in response to determining that the time of day associated with the network event is past a pre-determined time of day, adjusting the day of week to be a subsequent day of week;
determine, using the details of the network event and based on a machine learning model trained on historical network event data, a predicted verification delay for verifying the network event by a recipient of the network event, wherein training the machine learning model comprises: preprocessing the historical network event data including non-numeric categorical data and continuous data to encode the non-numeric categorical data into encoded data and to convert the continuous data into converted numerical data, and wherein the machine learning model is trained based on the encoded data and the converted numerical data;
obtain a set of relevant non-verification days for the network event based at least in part on a comparison between the historical network event data and a set of past non-verification days, wherein the set of relevant non-verification days correspond to days on which verifying the network event by the recipient cannot occur;
in response to determining that one or more non-verification days from the set of relevant non-verification days occur within a time period from the day of week and over the predicted verification delay, increase the predicted verification delay based on the one or more non-verification days;
provide the predicted verification delay to one or more downstream services for performing a planned operation based on an estimated verification delay, the predicted verification delay corresponding to the estimated verification delay, wherein the downstream services are configured to perform processing resource planning based on the estimated verification delay, thereby resulting in a reduction of processing resource utilization upon verification of the network event; and
when an actual verification delay extends beyond the predicted verification delay, flagging the network event and triggering an alert for intervention.
12 . The electronic device of claim 11 , wherein the processor is further configured to, prior to determination of the predicted verification delay:
determine a baseline verification delay for the network event based on a segment of the historical network event data; and
in response to the predicted verification delay being greater than the baseline verification delay, adjust the day of week of the network event to a next day that is not a non-verification day.
13 . The electronic device of claim 11 , wherein the processor is configured to increase the predicted verification delay by:
in response to the one or more non-verification days coinciding with the predicted verification delay, increase the predicted verification delay by one day.
14 . The electronic device of claim 11 , wherein the processor is configured to increase the predicted verification delay based on the one or more non-verification days by extending the predicted verification delay by one day.
15 . The electronic device of claim 11 , wherein the processor is further configured to update one or more attributes of the network event based on the predicted verification delay.
16 . The electronic device of claim 11 , wherein the processor is further configured to:
receive, from a remote server, reconciliation data associated with the network event;
determine a difference between the predicted verification delay and the actual verification delay indicated by the reconciliation data; and
re-train the machine learning model based on the difference.
17 . The electronic device of claim 11 , wherein the processor is configured to obtain the set of relevant non-verification days by filtering the set of past non-verification days by at least a country associated with the network event.
18 . The electronic device of claim 11 , wherein the set of past non-verification days includes past holidays.
19 . The electronic device of claim 11 , wherein the one or more downstream services comprise at least one of a safeguarding service, a resource management service, an intraday alerting service, or a foreign resource service.
20 . A non-transitory computer-readable medium comprising:
computer-readable instructions that, when executed by a processor, cause the processor to perform one or more operations comprising:
obtaining details of a network event, the details for the network event indicating a day of week associated with the network event and a time of day associated with the network event;
in response to determining that the time of day associated with the network event is past a pre-determined time of day, adjusting the day of week to be a subsequent day of week;
determining, using the details of the network event and based on a machine learning model trained on historical network event data, a predicted verification delay for verifying the network event by a recipient of the network event, wherein training the machine learning model comprises: preprocessing the historical network event data including non-numeric categorical data and continuous data to encode the non-numeric categorical data into encoded data and to convert the continuous data into converted numerical data, and wherein the machine learning model is trained based on the encoded data and the converted numerical data;
obtaining a set of relevant non-verification days for the network event based at least in part on a comparison between the historical network event data and a set of past non-verification days, wherein the set of relevant non-verification days correspond to days on which verifying the network event by the recipient cannot occur;
in response to determining that one or more non-verification days from the set of relevant non-verification days occur within a time period from the day of week and over the predicted verification delay, increasing the predicted verification delay based on the one or more non-verification days; and
providing the predicted verification delay to one or more downstream services for performing a planned operation based on an estimated verification delay, the predicted verification delay corresponding to the estimated verification delay, wherein the downstream services are configured to perform processing resource planning based on the estimated verification delay, thereby resulting in a reduction of processing resource utilization upon verification of the network event; and
when an actual verification delay extends beyond the predicted verification delay, flagging the network event and triggering an alert for intervention.