Systems and methods for generating structured datasets for predicting bond price with improved accuracy

1 . A non-transitory machine-readable medium, comprising executable instructions that, when executed by a processing system including a processor, facilitate performance of operations, the operations comprising:

storing collection of bond data identified and sorted based on a set of extractable features;

preprocessing the bond data by extracting the set of extractable features and partitioning the bond data into a plurality of bond clusters, wherein each volume of the bond data and the plurality of bond clusters correspond to daily trading volume that cannot be manually handled;

constructing a price function of each bond data contained in each of a plurality of bond clusters including a target cluster during a predetermined time window;

training a machine learning model by:

providing the target cluster to the machine learning model, wherein the target cluster comprises a bond dataset having at least one common data feature;

generating a price prediction result of the bond dataset by using the machine learning model;

determining occurrence of an error in the price prediction result based on a comparison against the price function of each bond data in the bond dataset;

upon the occurrence of the error in the price prediction result, identifying a cause that has resulted in the error in the price prediction result, wherein the cause includes a presence of one or more bond data in the bond dataset having a data feature deviated from remaining bond data in the bond dataset during the predetermined time window, wherein the identifying the cause further comprises identifying the cause by:

computing a first error percentage and a first correlation coefficient with respect to all of bond data in the bond dataset having the occurrence of the error in the price prediction result, wherein computing the first correlation coefficient further comprises computing the first correlation coefficient with respect to each pair of bond data and correlation of pricing trends of each pair of bond data;

determining whether the first error percentage exceeds a predetermined bond error threshold and the first correlation coefficient exceeds a predetermined bond correlation threshold; and

generating a feedback output identifying a resultant bond data in the bond dataset that have been determined to exceed the predetermined bond error threshold, the predetermined bond correlation threshold or both; and

based on the feedback output, reclustering the target cluster by including the resultant bond data and excluding the remaining bond data other than the resultant bond data resulting in a reclustered target cluster;

generating structured datasets including the reclustered target cluster;

providing the structured datasets to the machine learning model to generate a bond price prediction for a bond based on the structured datasets;

obtaining the a bond price prediction for the bond based on the structured datasets from the machine learning model;

determining a presence or absence of an error in the bond price prediction for the bond based on the structured datasets;

repeating the constructing, the training, the reclustering and the generating of the structured datasets to obtain adjusted structured datasets and providing the adjusted structured datasets to the machine learning model until the absence of the error in the bond price prediction is determined, wherein the repeating of the constructing, the training, the reclustering and the generating are automated and performed by a machine including the processing system; and

purchasing the bond based on the bond price prediction.

2 . The non-transitory machine-readable medium of claim 1 , wherein the training the machine learning model further comprises training a Hidden Markov Model algorithm resulting in a trained Hidden Markov Model.

3 . The non-transitory machine-readable medium of claim 2 , wherein the operations further comprise providing the generated structured datasets to the trained Hidden Markov Model algorithm.

4 . The non-transitory machine-readable medium of claim 1 , wherein the operations further comprise constructing one or more new clusters including the reclustered target cluster, and wherein the generating the structured datasets further comprises generating the structured datasets including the one or more new clusters.

5 . The non-transitory machine-readable medium of claim 4 , wherein the training the machine learning model further comprises:

computing a second error percentage and a second correlation coefficient with respect to the target cluster;

determining whether the second error percentage exceeds a predetermined cluster error threshold and the second correlation coefficient exceeds a predetermined cluster correlation threshold; and

generating another feedback output identifying that the target cluster has been determined to exceed the predetermined cluster error threshold, the predetermined cluster correlation threshold or both.

6 . The non-transitory machine-readable medium of claim 1 , wherein the computing the first correlation coefficient further comprises computing a Pearson correlation coefficient with respect to each pair of bonds included in the target cluster.

7 . The non-transitory machine-readable medium of claim 6 , wherein the training the machine learning model further comprises:

transforming the Pearson correlation coefficient (ρ) into a distance metric (d) using the following equation:

d

=

(

3

-

ρ

)

/

2

;

and

based on the distance metric (d), determining the remaining bond data to be removed from the target cluster or to be placed in a new cluster.

8 . A method, comprising:

storing, by a processing system including a processor, collection of bond data identified and sorted based on a set of extractable features;

preprocessing, by the processing system, the bond data by extracting the set of extractable features and partitioning the bond data into a plurality of bond clusters, wherein each volume of the bond data and the plurality of bond clusters correspond to daily trading volume that cannot be manually handled;

constructing, by the processing system, a price function of each bond contained in the plurality of bond clusters including a target cluster during a predetermined time window;

training, by the processing system, a machine learning model by:

providing the target cluster to the machine learning model, wherein the target cluster comprises a bond dataset having at least one common data feature;

generating a price prediction result of the bond dataset by using the machine learning model;

determining occurrence of an error in the price prediction result based on a comparison against the price function of each bond data in the bond dataset;

upon the occurrence of the error in the price prediction result, identifying a cause that has resulted in the error in the price prediction result, wherein the cause includes a presence of one or more bond data in the bond dataset having a data feature deviated from remaining bond data in the bond dataset during the predetermined time window, wherein the identifying the cause comprises identifying the cause by:

computing a first error percentage and a first correlation coefficient with respect to all of bond data in the bond dataset having the occurrence of the error in the price prediction result, wherein computing the first correlation coefficient further comprises computing the first correlation coefficient with respect to each pair of bond data and correlation of pricing trends of each pair of bond data; and

computing a second error percentage and a second correlation coefficient with respect to the target cluster;

determining whether the first error percentage exceeds a predetermined bond error threshold and the first correlation coefficient exceeds a predetermined bond correlation threshold;

determining whether the second error percentage exceeds a predetermined cluster error threshold and the second correlation coefficient exceeds a predetermined cluster correlation threshold; and

generating a feedback output identifying a resultant bond data in the bond dataset that have been determined to exceed the predetermined bond error threshold, the predetermined bond correlation threshold or both; and

based on the feedback output, reclustering, by the processing system, the target cluster by including the resultant bond data and excluding the remaining bond data other than the resultant bond data resulting in a reclustered target cluster, and generating one or more new clusters including the reclustered target cluster;

generating, by the processing system, structured datasets including the reclustered target cluster and the one or more new clusters;

providing the structured datasets to the machine learning model to generate a bond price prediction for a bond based on the structured datasets;

obtaining generating a bond price prediction for the bond based on the structured datasets from the machine learning model;

determining a presence or absence of an error in the bond price prediction for the bond based on the structured datasets;

repeating the constructing, the training, the reclustering and the generating of the structured datasets to obtain adjusted structured datasets and providing the adjusted structured datasets to the machine learning model until the absence of the error in the bond price prediction is determined, wherein the repeating of the constructing, the training, the reclustering and the generating are automated and performed by a machine including the processing system; and

purchasing a bond based on the bond price prediction.

9 . The method of claim 8 , wherein the training the machine learning model further comprises training a Hidden Markov Model algorithm.

10 . The method of claim 8 , further comprises wherein the computing the second error percentage further comprises computing an average error percentage of each bond included in the target cluster and the computing the second correlation coefficient further comprises computing an average correlation coefficient of each bond included in the target cluster.

11 . The method of claim 8 , wherein the computing the first correlation coefficient further comprises computing a Pearson correlation coefficient with respect to each pair of bonds included in the target cluster.

12 . The method of claim 11 , wherein the training the machine learning model further comprises:

transforming the Pearson correlation coefficient (p) into a distance metric (d); and

based on the distance metric (d), determining the resultant bond data in the bond dataset to be removed from the target cluster or to be placed in the one or more new clusters.

13 . A device, comprising:

a processing system including a processor; and

a memory that stores executable instructions that, when executed by the processing system, facilitate performance of operations, the operations comprising:

storing collection of bond data identified and sorted based on a set of extractable features;

preprocessing the bond data by extracting the set of extractable features and partitioning the bond data into a plurality of bond clusters, wherein each volume of the bond data and the plurality of bond clusters correspond to daily trading volume that cannot be manually handled;

constructing a price function of each bond data contained in a target cluster during a predetermined time window;

training a machine learning model by:

providing the target cluster to the machine learning model, wherein the target cluster comprises at least three bond data having at least one common data feature, wherein respective price functions of a first bond data and a second bond data among the at least three bond data follow substantially consistent trends;

generating a price prediction result of the at least three bond data by using the machine learning model;

upon an occurrence of an error in the price prediction result, identifying a cause that has resulted in the error in the price prediction result, wherein the cause includes a presence of one or more bond data having a data feature deviated from remaining bond data during the predetermined time window, wherein the identifying the cause comprises identifying the cause by:

computing a first correlation coefficient between the first and the second bond data with respect to correlation of pricing trends of the first and the second bond data;

computing a second correlation coefficient between the first bond data and a third bond data with respect to correlation of pricing trends of the first and the third bond data, or between the second bond data and the third bond data with respect to correlation of pricing trends of the second and the third bond data;

determining that the first correlation coefficient and the second correlation coefficient exceed a predetermined bond correlation threshold, respectively;

generating a feedback output identifying the third bond data that has been determined to exceed the predetermined bond correlation threshold;

based on the feedback output, removing the third bond data from the target cluster; and

generating structured datasets including the target cluster having the first bond data and the second bond data and a new cluster having the third bond data;

providing the structured datasets to the machine learning model to generate a bond price prediction for a bond based on the structured datasets;

obtaining a bond priced prediction for the bond based on the structure datasets;

determining a presence or absence of an error in the bond price prediction for the bond based on the structured datasets;

repeating the constructing, the training, the removing and the generating of the structured datasets to obtain adjusted structured datasets and providing the adjusted structured datasets to the machine learning model until the absence of the error in the bond price prediction is determined, wherein the repeating of the constructing, the training, the removing and the generating are automated and performed by a machine including the processing system; and

purchasing the bond based on the bond price prediction.

14 . The device of claim 13 , wherein the machine learning model comprises a Hidden Markov Model algorithm.

15 . The device of claim 14 , wherein the operations further comprise:

providing the structured datasets to the Hidden Markov Model algorithm; and

generating price prediction results of the target cluster and the new cluster.

16 . The device of claim 13 , wherein the training the machine learning model further comprises:

determining occurrence of an error in the price prediction result based on a comparison against the price function of each bond;

computing an error percentage with respect to each bond having the occurrence of the error in the price prediction result; and

determining whether the computed error percentage exceeds a predetermined bond error threshold; and

wherein the generating the feedback output further comprises generating the feedback output identifying all of bonds that have been determined to exceed the predetermined bond error threshold.

17 . The device of claim 13 , wherein:

the computing the first correlation coefficient further comprises computing a Pearson correlation coefficient with respect to the first bond data and the second bond data; and

the computing the second correlation coefficient further comprises computing the Pearson correlation coefficient with respect to the first bond data and the third bond data or with respect to the second bond data and the third bond data.

18 . The device of claim 17 , wherein the training the machine learning model further comprises transforming the Pearson correlation coefficient (p) into a distance metric (d); and

based on the distance metric (d), determining a new cluster to place the removed third bond data.

19 . The device of claim 13 , wherein the first correlation coefficient is greater than the second correlation coefficient.

20 . The device of claim 18 , wherein the distance metric (d) of the first and the second bond data is smaller than the distance metric (d) of the first and the third bond data and the distance metric (d) of the second and the third bond data.