| CPC G06Q 20/389 (2013.01) [G06Q 20/401 (2013.01)] | 12 Claims |
|
1. A method comprising:
training machine learning algorithms using a training dataset to generate a machine learning classification model, the training dataset constructed based on historical transaction data comprising historical transaction attributes for a plurality of historical transactions, wherein the machine learning classification model is configured to, provided as an input a set of transaction attributes associated with a given transaction, output, for the given transaction, a prediction indicating a temporal class among a plurality of temporal classes including an early class, an on time class, and a delay class;
training a first machine learning model using the training dataset to generate a first machine learning regression model, wherein the first machine learning regression model is configured to, provided as an input the set of transaction attributes associated with the given transaction and the temporal class of the given transaction, output a prediction of tranche delay days for the given transaction, wherein the first machine learning model includes one of a convolutional neural network, a recurrent neural network, or a deep neural network;
training a second machine learning model using the training dataset to generate a second machine learning regression model, wherein the second machine learning regression model is configured to, provided as an input the set of transaction attributes associated with the given transaction and the temporal class of the given transaction, output a prediction of a tranche count for the given transaction, wherein the second machine learning model includes one of a convolutional neural network, a recurrent neural network, or a deep neural network;
receiving, by a processor, a data record for an original transaction, wherein the data record comprises original transaction attributes including an amount of the original transaction, an original date upon which the original transaction is anticipated, and a party, a counterparty, or both to the original transaction;
inputting, by the processor, the original transaction attributes into the machine learning classification model to obtain a prediction of a temporal class for the original transaction, the temporal class comprising one from among the early class and the delay class;
inputting, by the processor into the first machine learning regression model, the original transaction attributes and information indicating the temporal class for the original transaction, to obtain a prediction of first tranche delay days, the first tranche delay days being a predicted temporal deviation from the original date for the original transaction, the temporal class having been obtained from the machine learning classification model;
inputting, by the processor, into the second machine learning regression model, the original transaction attributes and the information indicating the temporal class for the original transaction, to obtain a prediction of a tranche count for the original transaction, the temporal class having been obtained from the machine learning classification model;
rebuilding, by the processor, the original transaction as one or more future transactions, the rebuilding comprising:
determining at least one updated transaction attribute respectively corresponding to each of the one or more future transactions based on at least the temporal class output by the machine learning classification model for the original transaction, the first tranche delay days output by the first machine learning regression model for the original transaction, and the tranche count output by the second machine learning regression model for the original transaction, the at least one updated transaction attribute comprising one of an updated transaction amount with respect to the original transaction, an updated date upon which the updated transaction amount is anticipated, or both, and
generating an updated data record with regard to the original transaction so that the updated data record comprises, for each of the one or more future transactions, the at least one updated transaction attribute and at least one original transaction attribute among the original transaction attributes;
calculating, by the processor, a cash position based on the one or more future transactions; and
outputting, by the processor, the updated data record respectively corresponding to each of the one or more future transactions, the cash position, or both to an output device,
wherein the original transaction is processed as the one or more future transactions.
|