Machine learning techniques for generating disease prediction utilizing cross-temporal semi-structured input data
Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing predictive data analysis using semi-structured input data. Certain embodiments of the present invention utilize systems, methods, and computer program products that perform predictive data analysis using semi-structured input data using at least one of techniques using inferred codified fields and temporally-arranged codified fields.
1. A computer-implemented method comprising:
identifying, by one or more processors, a non-codified input data object that is generated by a respective data ingestion source of a plurality of data ingestion sources and is associated with a respective temporal marker of a plurality of temporal markers;
generating, by the one or more processors, an inferred record set comprising one or more inferred records based at least in part on the non-codified input data object, wherein: (i) an inferred record of the inferred record set is associated with a record field set comprising a plurality of record fields, and (ii) the plurality of record fields for the inferred record comprises an ingestion source identifier field, a source-specific data type identifier field, and a data value field;
generating, by the one or more processors, a discretized data value code for the inferred record based at least in part on (i) a subrange mapping scheme for a source-specific data type identifier code associated with the source-specific data type identifier field and (ii) the data value field for the inferred record;
generating, by the one or more processors, an inferred codified field set comprising a plurality of inferred codified fields based at least in part on the one or more inferred records, wherein: (i) an inferred codified field of the plurality of inferred codified fields is associated with the inferred record and is generated based at least in part on the plurality of record fields for the inferred record, and (ii) the inferred codified field comprises an ingestion source identifier code that is generated based at least in part on the ingestion source identifier field for the inferred record, a source-specific data type identifier code associated with the source-specific data type identifier field for the inferred record, and the discretized data value code;
generating, by the one or more processors, a temporally-arranged codified field set comprising a temporal arrangement of a group of input codified fields, wherein: (i) the group of input codified fields comprises the inferred codified field set and one or more input codified field sets, (ii) an input codified field of the one or more input codified field sets is associated with a corresponding temporal marker, and (iii) the corresponding temporal marker for the inferred codified field is generated based at least in part on the respective temporal marker for the non-codified input data object that is used to generate the inferred codified field;
generating, by the one or more processors and using a temporally encoded prediction machine learning model, a predictive output based at least in part on the temporally-arranged codified field set; and
performing, by the one or more processors, one or more prediction-based actions based at least in part on the temporally-arranged codified field set.
2. The computer-implemented method of claim 1 , wherein, given T temporal markers that are associated with a temporal marker sequence:
the temporally-arranged codified field set comprises a subset sequence of T codified field subsets;
a codified field subset of the temporally-arranged codified field set is associated with an associated temporal marker in the temporal marker sequence; and
the subset sequence is determined based at least in part on the temporal marker sequence.
3. The computer-implemented method of claim 1 , wherein generating the discretized data value code for the particular inferred codified field comprises:
identifying a total value range associated with the source-specific data type identifier code that is associated with the inferred record for the inferred codified field;
determining S value subranges for the total value range, wherein S is defined by a range segmentation count hyperparameter for the source-specific data type identifier code that is associated with the inferred record for the inferred codified field;
determining a selected value subrange from the S value subranges for the inferred codified field based at least in part on the data value field for the inferred record for the particular inferred codified field; and
generating the discretized data value code based at least in part on the selected value subrange.
4. The computer-implemented method of claim 3 , wherein the subrange mapping scheme maps each value subrange in the S value subranges to a respective subrange coding symbol.
5. The computer-implemented method of claim 4 , further comprising:
identifying the respective subrange coding symbol for the selected value subrange based at least in part on the subrange mapping scheme; and
generating the discretized data value code based at least in part on the respective subrange coding symbol.
6. The computer-implemented method of claim 1 , wherein the temporally encoded prediction machine learning model comprises a recurrent neural network machine learning model.
7. The computer-implemented method of claim 1 , wherein the temporally encoded prediction machine learning model comprises an encoder machine learning model that is trained as part of an encoder-decoder machine learning architecture.
8. The computer-implemented method of claim 1 , further comprising:
generating a natural language embedding vector for a natural language field associated with the inferred record; and
generating the data value field based at least in part on a vector distribution measure for the natural language embedding vector.
9. A system comprising memory and one or more processors communicatively coupled to the memory, the one or more processors configured to:
identify a non-codified input data object that is generated by a respective data ingestion source of a plurality of data ingestion sources and is associated with a respective temporal marker of a plurality of temporal markers;
generate an inferred record set comprising one or more inferred records based at least in part on the non-codified input data object, wherein: (i) an inferred record of the inferred record set is associated with a record field set comprising a plurality of record fields, and (ii) the plurality of record fields for the inferred record comprises an ingestion source identifier field, a source-specific data type identifier field, and a data value field;
generate a discretized data value code for the inferred record based at least in part on (i) a subrange mapping scheme for a source-specific data type identifier code associated with the source-specific data type identifier field and (ii) the data value field for the inferred record;
generate an inferred codified field set comprising a plurality of inferred codified fields based at least in part on the one or more inferred records, wherein: (i) an inferred codified field of the plurality of inferred codified fields is associated with the inferred record and is generated based at least in part on the plurality of record fields for the inferred record, and (ii) the inferred codified field comprises an ingestion source identifier code that is generated based at least in part on the ingestion source identifier field for the inferred record, a source-specific data type identifier code associated with the source-specific data type identifier field for the respective inferred record, and the discretized data value code;
generate a temporally-arranged codified field set comprising a temporal arrangement of a group of input codified fields, wherein: (i) the group of input codified fields comprises the inferred codified field set and one or more input codified field sets, (ii) an input codified field of the one or more input codified field sets is associated with a corresponding temporal marker, and (iii) the corresponding temporal marker for the inferred codified field is generated based at least in part on the respective temporal marker for the non-codified input data object that is used to generate the inferred codified field;
generate, using a temporally encoded prediction machine learning model, a predictive output based at least in part on the temporally-arranged codified field set; and
perform one or more prediction-based actions based at least in part on the temporally-arranged codified field set.
10. The system of claim 9 , wherein, given T temporal markers that are associated with a temporal marker sequence:
the temporally-arranged codified field set comprises a subset sequence of T codified field subsets;
a codified field subset of the temporally-arranged codified field set is associated with an associated temporal marker in the temporal marker sequence; and
the subset sequence is determined based at least in part on the temporal marker sequence.
11. The system of claim 9 , wherein the one or more processors are further configured to:
identify a total value range associated with the source-specific data type identifier code that is associated with the inferred record for the inferred codified field;
determine S value subranges for the total value range, wherein Sis defined by a range segmentation count hyperparameter for the source-specific data type identifier code that is associated with the inferred record for the inferred codified field;
determine a selected value subrange from the S value subranges for the inferred codified field based at least in part on the data value field for the inferred record for the inferred codified field; and
generate the discretized data value code based at least in part on the selected value subrange.
12. The system of claim 11 , wherein the subrange mapping scheme maps each value subrange in the S value subranges to a respective subrange coding symbol.
13. The system of claim 12 , wherein the one or more processors are further configured to:
identify the respective subrange coding symbol for the selected value subrange based at least in part on the subrange mapping scheme; and
generate the discretized data value code based at least in part on the respective subrange coding symbol.
14. The system of claim 9 , wherein the temporally encoded prediction machine learning model comprises a recurrent neural network machine learning model.
15. The system of claim 9 , wherein the temporally encoded prediction machine learning model comprises an encoder machine learning model that is trained as part of an encoder-decoder machine learning architecture.
16. The system of claim 9 , wherein the one or more processors are further configured to:
generate a natural language embedding vector for a natural language field associated with the inferred record; and
generate the data value field based at least in part on a vector distribution measure for the natural language embedding vector.
17. One or more non-transitory computer-readable storage media comprising instructions that, when executed by one or more processors, cause the one or more processors to:
identify a non-codified input data object that is generated by a respective data ingestion source of a plurality of data ingestion sources and is associated with a respective temporal marker of a plurality of temporal markers;
generate an inferred record set comprising one or more inferred records based at least in part on the non-codified input data object, wherein: (i) an inferred record of the inferred record set is associated with a record field set comprising a plurality of record fields, and (ii) the plurality of record fields for the inferred record comprises an ingestion source identifier field, a source-specific data type identifier field, and a data value field;
generate a discretized data value code for the inferred record based at least in part on (i) a subrange mapping scheme for a source-specific data type identifier code associated with the source-specific data type identifier field and (ii) the data value field for the inferred record;
generate an inferred codified field set comprising a plurality of inferred codified fields based at least in part on the one or more inferred records, wherein: (i) an inferred codified field of the plurality of inferred codified fields is associated with the inferred record and is generated based at least in part on the plurality of record fields for the inferred record, and (ii) the inferred codified field comprises an ingestion source identifier code that is generated based at least in part on the ingestion source identifier field for the inferred record, a source-specific data type identifier code associated with the source-specific data type identifier field for the inferred record, and the discretized data value code;
generate a temporally-arranged codified field set comprising a temporal arrangement of a group of input codified fields, wherein: (i) the group of input codified fields comprises the inferred codified field set and one or more input codified field sets, (ii) an input codified field of the one or more input codified field sets is associated with a corresponding temporal marker, and (iii) the corresponding temporal marker for the inferred codified field is generated based at least in part on the respective temporal marker for the non-codified input data object that is used to generate the inferred codified field;
generate, using a temporally encoded prediction machine learning model, a predictive output based at least in part on the temporally-arranged codified field set; and
perform one or more prediction-based actions based at least in part on the temporally-arranged codified field set.
18. The one or more non-transitory computer-readable storage media of claim 17 , wherein, given T temporal markers that are associated with a temporal marker sequence:
the temporally-arranged codified field set comprises a subset sequence of T codified field subsets;
a codified field subset of the temporally-arranged codified field set is associated with an associated temporal marker in the temporal marker sequence; and
the subset sequence is determined based at least in part on the temporal marker sequence.
19. The one or more non-transitory computer-readable storage media of claim 17 , wherein the instructions further cause the one or more processors to:
identify a total value range associated with the source-specific data type identifier code that is associated with the inferred record for the inferred codified field;
determine S value subranges for the total value range, wherein S is defined by a range segmentation count hyperparameter for the source-specific data type identifier code that is associated with the inferred record for the inferred codified field;
determine a selected value subrange from the S value subranges for the inferred codified field based at least in part on the data value field for the inferred record for the inferred codified field; and
generate the discretized data value code based at least in part on the selected value subrange.
20. The one or more non-transitory computer-readable storage media of claim 19 , wherein the subrange mapping scheme maps each value subrange in the S value subranges to a respective subrange coding symbol.