Systems and methods for automated generation classifiers
Systems and methods to automatically generate classifiers are provided. A labeled dataset is initially received. The dataset may be for a positive class, or may be a negative for a class, or a false positive class. N features that are predictive for the class (or false positive or the negative class) are identified. These features are combined within a classifier dictionary. Medical records received may be processed in order to be machine readable. Features within the medical records are identified and are compared against the dictionary of classifiers. Matches indicate classes within the medical record. The classifier dictionary may be periodically updated in response to insufficient classification accuracy, or when new data becomes available.
1. A computerized method for generating classifiers implemented using an automated classifier computer system comprising processor in communication with a memory, the method comprising:
receiving a first labeled dataset associated with positive medical diagnoses, a second labeled dataset associated with false positive medical diagnoses, and a third labeled dataset associated with negative medical diagnoses;
determining N features predictive of a positive medical diagnosis using the first labeled dataset, M features predictive of a false positive medical diagnosis using the second labeled dataset, and P features predictive of a negative medical diagnosis using the third labeled dataset;
generating a first feature dictionary by associating the N features with a plurality of medical code classifiers;
generating a second feature dictionary by associating the M features with the plurality of medical code classifiers;
generating a third feature dictionary by associating the P features with the plurality of medical code classifiers;
generating a master feature dictionary for evaluating healthcare documents for code-able events by combining the first feature dictionary, the second feature dictionary, and the third feature dictionary;
receiving a digital medical record from an electronic records system;
identifying features in the digital medical record in a backend processor; and
determining at least one classifier of the plurality of medical code classifiers to assign to the digital medical record by utilizing the master feature dictionary.
2. The method of claim 1 , wherein finding N features includes mutual information feature selection methods.
3. The method of claim 1 , wherein the first labeled dataset includes an evaluation dataset and a training dataset.
4. The method of claim 3 , wherein finding N features uses the training dataset.
5. The method of claim 3 , further comprising identifying test features in the evaluation dataset, and matching the test features to the N features predictive of a positive medical diagnosis to determine at least one test classifier.
6. The method of claim 5 , further comprising comparing the at least one test classifier to at least one known classifier for the evaluation dataset.
7. The method of claim 6 , further comprising updating the N features if the at least one test classifier does not match the at least one known classifier.
8. The method of claim 1 , further comprising receiving, at an interface of the automated classifier computer system, the first labeled dataset associated with positive medical diagnoses, the second labeled dataset associated with false positive medical diagnoses, and the third labeled dataset associated with negative medical diagnoses.
9. An automated classifier computer system comprising:
a feature selector embodied on an analytics layer of a server system configured to:
receive a first labeled dataset associated with positive medical diagnoses, a second labeled dataset associated with false positive medical diagnoses, and a third labeled dataset associated with negative medical diagnoses;
determine N features predictive of a positive medical diagnosis using the first labeled dataset, M features predictive of a false positive medical diagnosis using the second labeled dataset, and P features predictive of a negative medical diagnosis using the third labeled dataset;
generate a first feature dictionary by associating the N features with a plurality of medical code classifiers;
generate a second feature dictionary by associating the M features with the plurality of medical code classifiers;
generate a third feature dictionary by associating the P features with the plurality of medical code classifiers; and
generate a master feature dictionary for evaluating healthcare documents for code-able events by combining the first feature dictionary, the second feature dictionary, and the third feature dictionary; and
a classifier embodied on a backend processor configured to:
receive a digital medical record from an electronic records system;
identify features in the digital medical record; and
determine at least one classifier of the plurality of medical code classifiers to assign to the digital medical record by utilizing the master feature dictionary.
10. The system of claim 9 , wherein finding N features includes mutual information feature selection methods.
11. The system of claim 9 , wherein the first labeled dataset includes an evaluation dataset and a training dataset.
12. The system of claim 11 , wherein finding N features uses the training dataset.
13. The system of claim 11 , wherein the classifier is further configured to identify test features in the evaluation dataset, and match the test features to the N features predictive of a positive medical diagnosis to determine at least one test classifier.
14. The system of claim 13 , wherein the classifier is further configured to compare the at least one test classifier to at least one known classifier for the evaluation dataset.
15. The system of claim 14 , wherein the feature selector is further configured to update the N features if the at least one test classifier does not match the at least one known classifier.