System, method, and computer program product for multivariate event prediction using multi-stream recurrent neural networks

A system for multivariate event prediction, the system comprising a server including at least one hardware processor for executing program instructions, the server being programmed and/or configured to:

• receive event data associated with a plurality of events occurring in a sample time period, the sample time period comprising a plurality of days and each day comprising a plurality of subperiods;
• generate a plurality of feature vectors for each day, wherein each feature vector is associated with a subperiod of a day, wherein each feature vector comprises a plurality of features based on the event data, wherein generating the plurality of feature vectors comprises aggregating, for each respective feature vector, a plurality of values of each feature during a subperiod associated with the respective feature vector;
• provide the plurality of feature vectors as inputs to a set of first recurrent neural network (RNN) models, wherein the set of first RNN models comprises a first plurality of nodes, wherein the first plurality of nodes comprises a plurality of subsets of nodes, wherein each subset of nodes is associated with a day and a first RNN model of the set of first RNN models, wherein each node of a subset of nodes is associated with a subperiod, wherein providing the plurality of feature vectors as inputs to the set of first RNN models comprises providing each feature vector associated with a subperiod as an input to a respective RNN node of a respective first RNN model;
• generate first outputs for each RNN node of the set of first RNN models, wherein generating each of the first outputs comprises generating a hidden layer feature vector for a hidden layer of each RNN node based on a feature vector that was provided as an input to a respective RNN node, wherein each of the first outputs comprises the hidden layer feature vector;
• merge the first outputs for each same subperiod of the plurality of subperiods to form aggregated time-series layers, wherein each of the aggregated time-series layers is associated with a subperiod;
• provide the aggregated time-series layers as an input to a second RNN model, wherein the second RNN model comprises a second plurality of nodes, wherein each node of the second plurality of nodes is associated with a subperiod, wherein providing the aggregated time-series layers as an input to the second RNN model comprises providing each of the aggregated time-series layers as an input to a respective RNN node;
• and
• generate final outputs for each RNN node of the second RNN model, wherein generating each of the final outputs comprises generating a final hidden layer feature vector for a hidden layer of each RNN node of the second RNN model based on one of the aggregated time-series layers that was provided as an input to a respective RNN node, wherein each of the final outputs comprises the final hidden layer feature vector.

A computer program product for multivariate event prediction, the computer program product comprising at least one non-transitory computer-readable medium including program instructions that, when executed by at least one processor, cause the at least one processor to:

• receive event data associated with a plurality of events occurring in a sample time period, the sample time period comprising a plurality of days and each day comprising a plurality of subperiods;
• generate a plurality of feature vectors for each day, wherein each feature vector is associated with a subperiod of a day, wherein each feature vector comprises a plurality of features based on the event data, wherein generating the plurality of feature vectors comprises aggregating, for each respective feature vector, a plurality of values of each feature during a subperiod associated with the respective feature vector;
• provide the plurality of feature vectors as inputs to a set of first recurrent neural network (RNN) models, wherein the set of first RNN models comprises a first plurality of nodes, wherein the first plurality of nodes comprises a plurality of subsets of nodes, wherein each subset of nodes is associated with a day and a first RNN model of the set of first RNN models, wherein each node of a subset of nodes is associated with a subperiod, wherein providing the plurality of feature vectors as inputs to the set of first RNN models comprises providing each feature vector associated with a subperiod as an input to a respective RNN node of a respective first RNN model;
• generate first outputs for each RNN node of the set of first RNN models, wherein generating each of the first outputs comprises generating a hidden layer feature vector for a hidden layer of each RNN node based on a feature vector that was provided as an input to a respective RNN node, wherein each of the first outputs comprises the hidden layer feature vector;
• merge the first outputs for each same subperiod of the plurality of subperiods to form aggregated time-series layers, wherein each of the aggregated time-series layers is associated with a subperiod;
• provide the aggregated time-series layers as an input to a second RNN model, wherein the second RNN model comprises a second plurality of nodes, wherein each node of the second plurality of nodes is associated with a subperiod, wherein providing the aggregated time-series layers as an input to the second RNN model comprises providing each of the aggregated time-series layers as an input to a respective RNN node;
• and
• generate final outputs for each RNN node of the second RNN model, wherein generating each of the final outputs comprises generating a final hidden layer feature vector for a hidden layer of each RNN node of the second RNN model based on one of the aggregated time-series layers that was provided as an input to a respective RNN node, wherein each of the final outputs comprises the final hidden layer feature vector.