| CPC G06N 3/08 (2013.01) | 22 Claims |
|
1. A computer implemented method of training a machine learning network to classify a data record, the method comprising:
receiving a plurality of data records relating to workplace incidents, each of the plurality of data records having a plurality of field entries;
processing the plurality of field entries by performing a first data filtering based on a temporal distribution of the workplace incidents followed by a second data filtering based at least on missing value ratios to identify at least a first predefined textual field type and a second data field type;
inputting at least a first portion of the plurality of field entries of the first predefined textual field type into a deep neural network (DNN);
inputting at least a second portion of the plurality of field entries of the second data field type into a different machine learning model (ML);
encoding, via the DNN, the first portion of the plurality of field entries of the first predefined textual field type to output a densely embedded contextual vector;
encoding, via the ML, ordered values representing the second portion of the plurality of field entries of the second data field type into a sparse vector;
concatenating the densely embedded contextual vector with the sparse vector to generate a representative vector of the plurality of data records;
inputting the representative vector as training inputs into a gradient-boosted classifier network to generate a classification of each data record; and
tuning a hyperparameter of the gradient-boosted classifier network by using the representative vector as a tuning input.
|