Systems and methods for classifying imbalanced data

1 . A computer-implemented method comprising:

receiving a request to generate a classification model, wherein the request indicates a quantifiable loss threshold to be satisfied by the classification model, and wherein the request further includes a data set including first data associated with a first characteristic and second data associated with a second characteristic;

generating an initial iteration of the classification model, wherein the initial iteration of the classification model is generated by randomly initializing a set of model coefficients;

generating a modified loss function, wherein the modified loss function is generated to identify quantifiable losses resulting from misclassification of data points associated with the data set, wherein the modified loss function includes a set of constant input loss values for false positive classifications and false negative classifications;

processing the data set through the initial iteration of the classification model to obtain an output, wherein the output includes a classification of the first data into majority data based on the first characteristic and the second data into minority data based on the second characteristic;

determining an initial quantifiable loss corresponding to the initial iteration of the classification model, wherein the initial quantifiable loss is determined according to the modified loss function and misclassifications identified from the output, and wherein the initial quantifiable loss does not satisfy the quantifiable loss threshold;

updating one or more model coefficients of the set of model coefficients to generate a new iteration of the classification model, wherein the one or more model coefficients are updated through an optimization algorithm;

continuously updating the data set as new data points are obtained, wherein different data points of the new data points are associated either with the first characteristic or the second characteristic;

processing the updated data set through the new iteration of the classification model to obtain a new output including at least re-classifications of subsets of the first data into the minority data and subsets of the second data into the majority data;

determining that a new quantifiable loss corresponding to the new iteration of the classification model satisfies the quantifiable loss threshold, wherein the new quantifiable loss is determined according to the modified loss function and new misclassifications identified from the new output; and

providing the new iteration of the classification model as a result of the quantifiable loss threshold being satisfied.

2 . The computer-implemented method of claim 1 , wherein the one or more model coefficients are updated by the optimization algorithm using a gradient descent algorithm.

3 . The computer-implemented method of claim 1 , wherein the data set includes imbalanced data.

4 . The computer-implemented method of claim 1 , wherein the modified loss function is generated by incorporating a loss weighing component into a loss function, and wherein the loss weighing component includes the set of constant input loss values.

5 . The computer-implemented method of claim 1 , wherein the one or more model coefficients are updated according to a learning rate, and wherein the learning rate is adjusted according to the initial quantifiable loss.

6 . The computer-implemented method of claim 1 , wherein the initial quantifiable loss corresponds to a monetary amount lost as a result of the misclassifications.

7 . The computer-implemented method of claim 1 , further comprising:

updating the set of constant input loss values based on a statistical analysis of misclassifications of different data points associated with different data sets; and

calculating the new quantifiable loss according to the updated set of constant input loss values.

8 . A system, comprising:

one or more processors; and

memory storing thereon instructions that, as a result of being executed by the one or more processors, cause the system to:

receive a request to generate a classification model, wherein the request indicates a quantifiable loss threshold to be satisfied by the classification model, and wherein the request further includes a data set including first data associated with a first characteristic and second data associated with a second characteristic;

generate an initial iteration of the classification model, wherein the initial iteration of the classification model is generated by randomly initializing a set of model coefficients;

generate a modified loss function, wherein the modified loss function is generated to identify quantifiable losses resulting from misclassification of data points associated with the data set, wherein the modified loss function includes a set of constant input loss values for false positive classifications and false negative classifications;

process the data set through the initial iteration of the classification model to obtain an output, wherein the output includes a classification of the first data into majority data based on the first characteristic and the second data into minority data based on the second characteristic;

determine an initial quantifiable loss corresponding to the initial iteration of the classification model, wherein the initial quantifiable loss is determined according to the modified loss function and misclassifications identified from the output, and wherein the initial quantifiable loss does not satisfy the quantifiable loss threshold;

update one or more model coefficients of the set of model coefficients to generate a new iteration of the classification model, wherein the one or more model coefficients are updated through an optimization algorithm;

continuously update the data set as new data points are obtained, wherein different data points of the new data points are associated either with the first characteristic or the second characteristic;

process the updated data set through the new iteration of the classification model to obtain a new output including at least re-classifications of subsets of the first data into the minority data and subsets of the second data into the majority data;

determine that a new quantifiable loss corresponding to the new iteration of the classification model satisfies the quantifiable loss threshold, wherein the new quantifiable loss is determined according to the modified loss function and new misclassifications identified from the new output; and

provide the new iteration of the classification model as a result of the quantifiable loss threshold being satisfied.

9 . The system of claim 8 , wherein the one or more model coefficients are updated by the optimization algorithm using a gradient descent algorithm.

10 . The system of claim 8 , wherein the data set includes imbalanced data.

11 . The system of claim 8 , wherein the modified loss function is generated by incorporating a loss weighing component into a loss function, and wherein the loss weighing component includes the set of constant input loss values.

12 . The system of claim 8 , wherein the one or more model coefficients are updated according to a learning rate, and wherein the learning rate is adjusted according to the initial quantifiable loss.

13 . The system of claim 8 , wherein the initial quantifiable loss corresponds to a monetary amount lost as a result of the misclassifications.

14 . The system of claim 8 , wherein the instructions further cause the system to:

update the set of constant input loss values based on a statistical analysis of misclassifications of different data points associated with different data sets; and

calculate the new quantifiable loss according to the updated set of constant input loss values.

15 . A non-transitory, computer-readable storage medium storing thereon executable instructions that, as a result of being executed by one or more processors of a computer system, cause the computer system to:

receive a request to generate a classification model, wherein the request indicates a quantifiable loss threshold to be satisfied by the classification model, and wherein the request further includes a data set including first data associated with a first characteristic and second data associated with a second characteristic;

generate an initial iteration of the classification model, wherein the initial iteration of the classification model is generated by randomly initializing a set of model coefficients;

generate a modified loss function, wherein the modified loss function is generated to identify quantifiable losses resulting from misclassification of data points associated with the data set, wherein the modified loss function includes a set of constant input loss values for false positive classifications and false negative classifications;

process the data set through the initial iteration of the classification model to obtain an output, wherein the output includes a classification of the first data into majority data based on the first characteristic and the second data into minority data based on the second characteristic;

determine an initial quantifiable loss corresponding to the initial iteration of the classification model, wherein the initial quantifiable loss is determined according to the modified loss function and misclassifications identified from the output, and wherein the initial quantifiable loss does not satisfy the quantifiable loss threshold;

update one or more model coefficients of the set of model coefficients to generate a new iteration of the classification model, wherein the one or more model coefficients are updated through an optimization algorithm;

continuously update the data set as new data points are obtained, wherein different data points of the new data points are associated either with the first characteristic or the second characteristic;

process the updated data set through the new iteration of the classification model to obtain a new output including at least re-classifications of subsets of the first data into the minority data and subsets of the second data into the majority data;

determine that a new quantifiable loss corresponding to the new iteration of the classification model satisfies the quantifiable loss threshold, wherein the new quantifiable loss is determined according to the modified loss function and new misclassifications identified from the new output; and

provide the new iteration of the classification model as a result of the quantifiable loss threshold being satisfied.

16 . The non-transitory, computer-readable storage medium of claim 15 , wherein the one or more model coefficients are updated by the optimization algorithm using a gradient descent algorithm.

17 . The non-transitory, computer-readable storage medium of claim 15 , wherein the data set includes imbalanced data.

18 . The non-transitory, computer-readable storage medium of claim 15 , wherein the modified loss function is generated by incorporating a loss weighing component into a loss function, and wherein the loss weighing component includes the set of constant input loss values.

19 . The non-transitory, computer-readable storage medium of claim 15 , wherein the one or more model coefficients are updated according to a learning rate, and wherein the learning rate is adjusted according to the initial quantifiable loss.

20 . The non-transitory, computer-readable storage medium of claim 15 , wherein the initial quantifiable loss corresponds to a monetary amount lost as a result of the misclassifications.

21 . The non-transitory, computer-readable storage medium of claim 15 , wherein the executable instructions further cause the computer system to:

update the set of constant input loss values based on a statistical analysis of misclassifications of different data points associated with different data sets; and

calculate the new quantifiable loss according to the updated set of constant input loss values.