Automated training of machine learning classification for patient missed care opportunities or late arrivals

1 . A system, comprising:

a processor that executes computer-executable components stored in a computer-readable memory, the computer-executable components comprising:

a receiver component that accesses a first set of annotated data candidates defined by a plurality of feature categories, wherein the annotated data candidates respectively pertain to medical patients that have scheduled medical appointments;

a training component that trains a machine learning classifier using a first training process on first feature vectors extracted from the first set to produce as output for each annotated data candidate, a classification label regarding attendance of the scheduled medical appointments of among three classification labels, thereby causing internal parameters of the machine learning classifier to become iteratively updated during the first training process, wherein the three classification labels include an on time arrival label, a late arrival label and a no arrival label;

a feature component that extracts numerical representations of the internal parameters of the machine learning classifier after completion of the first training process, provides the numerical representations as input to an importance score generation machine learning model, and executes the importance score generation machine learning model to generate respective importance scores for the feature categories; and

an execution component that identifies respective feature categories of the plurality of feature categories having importance scores that fail to satisfy a threshold value and removes the respective feature categories from the first set, thereby yielding a second set of annotated data candidates, and wherein the training component retrains the machine learning classifier using a second training process on second feature vectors extracted from the second set, thereby generating an optimized version of the machine learning classifier, wherein the second feature vectors having a reduced feature vector dimensionality relative to the first feature vectors.

2 . The system of claim 1 , wherein the machine learning classifier is an XGBoost model, and wherein the internal parameters include decision tree thresholds of the XGBoost model or decision tree locations of the XGBoost model.

3 . The system of claim 1 , wherein the feature categories include a patient demographics feature category, a patient medical record feature category, a weather feature category, and a traffic feature category.

4 . The system of claim 1 , wherein feature categories included in the second set comprise:

an appointment-date category, an allocation-date category, an admission-type category, a department category, a modality category, an age category, an education category, an employment category, a pathology category, an ethnicity category, an insurance category, a residence category, a traffic category, and a weather category.

5 . The system of claim 1 , wherein the plurality of feature categories corresponds to a filtered set of feature categories generated from an initial set of feature categories associated with the first set of annotated data candidates, and wherein the computer-executable components further comprise:

a preprocessing component that removes one or more initial feature categories from the initial set to generate the filtered set of feature categories, wherein the preprocessing component removes the one or more initial feature categories based on at least one of: a first determination that the one or more initial feature categories exhibit a variance level below a threshold variance level across the set of annotated data candidates, or a second determination that a threshold proportion of the set of annotated data candidates includes a null or unavailable value for the one or more initial features.

6 . The system of claim 1 , wherein the machine learning classifier is a first neural network model comprising one or more softmax functions, and wherein the internal parameters include weight matrices of neurons of the first neural network model.

7 . The system of claim 1 , wherein the machine learning classifier comprises a logistic regression architecture, and wherein the internal parameters include regression coefficients.

8 . The system of claim 1 , the computer-executable components comprising:

a validation component that determines one or more performance metrics for the machine learning classifier after the retraining, compares the one or more performance metrics to a predefined performance threshold, and in response to the one or more performance metrics failing to satisfy the predefined performance threshold, directs the feature component to determine updated importance scores and directs the training component to again retrain the machine learning classifier using a further updated set of annotated data candidates determined by the execution component until the one or more performance metrics satisfy the predefined performance threshold.

9 . The system of claim 1 , the computer-executable components comprising:

a validation component that determines one or more performance metrics of the machine learning classifier after the completion of the first training process, and wherein the execution component initiates removal of the respective feature categories and retraining in the second training process based on the one or more performance metrics failing to satisfy a performance threshold.

10 . The system of claim 1 , wherein the machine learning classifier comprises a neural network architecture having weight matrices and bias values as the internal parameters, and wherein the numerical representations comprise one or more scalars, vectors, matrices, or tensors representing the weight matrices and bias values.

11 . The system of claim 1 , wherein the machine learning classifier comprises a decision tree architecture including decision nodes and leaf nodes, and wherein the importance scores are generated based at least in part on relative locations of decision nodes within the decision tree architecture.

12 . The system of claim 1 , wherein, after retraining in the second training process, the execution component electronically deploys the optimized version of the machine learning classifier to classify newly encountered data candidates in an operational healthcare environment.

13 . A computer-implemented method, comprising:

accessing, by a device operatively coupled to a processor, a first set of annotated data candidates defined by a plurality of feature categories, wherein the annotated data candidates respectively pertain to medical patients that have scheduled medical appointments;

training, by the device, a machine learning classifier sing a first training process on first feature vectors extracted from the first set to produce as output for each annotated data candidate, a classification label regarding attendance of the scheduled medical appointments of among three possible classification labels, thereby causing internal parameters of the machine learning classifier to become iteratively updated during the first training process, wherein the three classification labels include an on time arrival label, a late arrival label and a no arrival label;

extracting, by the device, numerical representations of the internal parameters of the machine learning classifier after completion of the first training process;

providing, by the device, the numerical representations as input to an importance score generation machine learning model;

executing, by the device, the importance score generation machine learning model to generate respective importance scores for the feature categories;

removing, by the device, respective feature categories of the plurality of feature categories of the first set having importance scores that fail to satisfy a threshold value, thereby yielding a second set of annotated data candidates; and

retraining, by the device, the machine learning classifier using a second machine learning process on second feature vectors extracted from the second set, thereby generating an optimized version of the machine learning classifier, wherein the second feature vectors having a reduced feature vector dimensionality relative to the first feature vectors.

14 . The computer-implemented method of claim 13 , wherein the machine learning classifier is an XGBoost model, and wherein the internal parameters include decision tree thresholds of the XGBoost model or decision tree locations of the XGBoost model.

15 . The computer-implemented method of claim 14 , wherein the first training comprises splitting, by the system, the first set of annotated data candidates according to annotation based on Gini impurity.

16 . The computer-implemented method of claim 13 , wherein the feature categories include a patient demographics feature category, a patient medical record feature category, a weather feature category, and a traffic feature category.

17 . A non-transitory computer program product for facilitating automated training of machine learning classification for patient missed care opportunities or late arrivals, the computer program product comprising a computer-readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

access a first set of annotated data candidates defined by a plurality of feature categories, wherein the annotated data candidates respectively pertain to medical patients that have scheduled medical appointments;

train a machine learning classifier sing a first training process on first feature vectors extracted from the first set to produce as output for each annotated data candidate, a classification label regarding attendance of the scheduled medical appointments of among three possible classification labels, thereby causing internal parameters of the machine learning classifier to become iteratively updated during the first training process, wherein the three classification labels include an on time arrival label, a late arrival label and a no arrival label;

extract numerical representations of the internal parameters of the machine learning classifier after completion of the first training process;

provide the numerical representations as input to an importance score generation machine learning model;

execute the importance score generation machine learning model to generate respective importance scores for the feature categories;

remove respective feature categories of the plurality of feature categories of the first set having importance scores that fail to satisfy a threshold value, thereby yielding a second set of annotated data candidates; and

retrain the machine learning classifier using a second machine learning process on second feature vectors extracted from the second set, thereby generating an optimized version of the machine learning classifier, wherein the second feature vectors having a reduced feature vector dimensionality relative to the first feature vectors.

18 . The non-transitory computer program product of claim 17 , wherein the machine learning classifier is an XGBoost model, and wherein the internal parameters include decision tree thresholds of the XGBoost model or decision tree locations of the XGBoost model.

19 . The non-transitory computer program product of claim 17 , wherein the program instructions further cause the processor to:

render, on an electronic display, the respective importance scores for the feature categories.

20 . The non-transitory computer program product of claim 17 , wherein the program instructions further cause the processor to:

render, on an electronic display, an indication of the respective feature categories that fail to satisfy the threshold value.