Prediction algorithm based attribute data processing

1. A computer-implemented method that identifies an insurance risk prediction model that generates a more accurate risk prediction result for a respective insurance service risk assessment and that requires personal attribute information, the computer-implemented method comprising:

receiving, at a prediction server, an insurance service scenario comprising variable data of personal attribute information associated with at least one insurance user;

generating, at the prediction server, at least two candidate insurance risk prediction models comprising a first model and a second model, wherein the first model comprises a neural network, and wherein the second model is constructed by modeling at least one modeling target value using a gradient boosting decision tree (GBDT), and wherein the at least one modeling target value includes at least one of a loss ratio, a claim frequency, and a claim amount of the at least one insurance user;

training, at the prediction server, the first model to include a learned continuous feature vector using a training objective function, wherein the training of the first model comprises:

randomly initializing, at the prediction server, parameters of the first model,

generating, at the prediction server, N bins each comprising data of a different feature type included in the variable data of personal attribute information,

generating, at the prediction server, a first discrete feature vector comprising at least N elements corresponding to the N bins,

mapping, at the prediction server, the first discrete feature vector and at least a second discrete feature vector to generate a candidate continuous feature vector,

inputting, at the prediction server and into the first model, the candidate continuous feature vector representing the variable data of personal attribute information,

obtaining, at the prediction server, output of the first model indicating a risk score based on the candidate continuous feature vector, and

optimizing, at the prediction server, the training objective function associated with the first model based on the risk score, the parameters of the first model, and the at least one modeling target value;

generating, at the prediction server, a first function output of a Tweedie regression objective function of the GBDT using the variable data of personal attribute information as an input to the second model;

generating, at the prediction server, a second function output of the first model using the variable data of personal attribute information that was used as an input to the first model;

generating, at the prediction server, a first statistical indicator based on the first function output and the input to the second model;

generating, at the prediction server, a second statistical indicator based on the second function output and the input to the first model;

selecting, at the prediction server and for the insurance service scenario, the insurance risk prediction model that is more statistically significant from the at least two candidate insurance risk prediction models based on a comparison of the first statistical indicator and the second statistical indicator; and

generating, at the prediction server and for the insurance service scenario, an insurance risk prediction result of the at least one insurance user using the selected insurance risk prediction model.

2. The computer-implemented method of claim 1 , further comprising preprocessing the variable data of personal attribute information, wherein the preprocessing, at the prediction server, comprises at least one of:

setting a weight for each personal attribute of the variable data;

supplementing an incomplete value with a default value in the variable data;

determining a selection of repeated variable data; processing interaction effects between the variable data; or generating new variable data based on the variable data.

3. The computer-implemented method of claim 1 , wherein at least one of the at least two candidate insurance risk prediction models is a trained generalized linear model.

4. The computer-implemented method of claim 1 , wherein the GBDT uses a gamma regression objective function as an objective function, and wherein the at least one modeling target value using the GBDT includes a difference between an actual modeling target value and a predicted modeling target value.

5. The computer-implemented method of claim 1 , wherein the training objective function is optimized using a Stochastic Gradient Descent.

6. A non-transitory, computer-readable medium that identifies an insurance risk prediction model that generates a more accurate risk prediction result for a respective insurance service risk assessment and that requires personal attribute information, the non-transitory, computer-readable medium storing one or more instructions executable by a computer processor, included in a prediction server, to perform operations comprising:

receiving, at the prediction server, an insurance service scenario comprising variable data of personal attribute information associated with at least one insurance user;

generating, at the prediction server, at least two candidate insurance risk prediction models comprising a first model and a second model, wherein the first model comprises a neural network, and wherein the second model is constructed by modeling at least one modeling target value using a gradient boosting decision tree (GBDT), and wherein the at least one modeling target value includes at least one of a loss ratio, a claim frequency, and a claim amount of the at least one insurance user;

training, at the prediction server, the first model to include a learned continuous feature vector using a training objective function, wherein the training of the first model comprises:

randomly initializing, at the prediction server, parameters of the first model,

generating, at the prediction server, N bins each comprising data of a different feature type included in the variable data of personal attribute information,

generating, at the prediction server, a first discrete feature vector comprising at least N elements corresponding to the N bins,

mapping, at the prediction server, the first discrete feature vector and at least a second discrete feature vector to generate a candidate continuous feature vector,

inputting, at the prediction server and into the first model, the candidate continuous feature vector representing the variable data of personal attribute information,

obtaining, at the prediction server, output of the first model indicating a risk score based on the candidate continuous feature, and

optimizing, at the prediction server, the training objective function associated with the first model based on the risk score, the parameters of the first model, and the at least one modeling target value;

generating, at the prediction server, a first function output of a Tweedie regression objective function of the GBDT using the variable data of personal attribute information as an input to the second model;

generating, at the prediction server, a second function output of the first model using the variable data of personal attribute information that was used as an input to the first model;

generating, at the prediction server, a first statistical indicator based on the first function output and the input to the second model;

generating, at the prediction server, a second statistical indicator based on the second function output and the input to the first model;

selecting, at the prediction server and for the insurance service scenario, the insurance risk prediction model that is more statistically significant from the at least two candidate insurance risk prediction models based on a comparison of the first statistical indicator and the second statistical indicator; and

generating, at the prediction server and for the insurance service scenario, an insurance risk prediction result of the at least one insurance user using the selected insurance risk prediction model.

7. The non-transitory, computer-readable medium of claim 6 , further comprising one or more instructions to preprocess the variable data of personal attribute information, wherein the preprocessing, at the prediction server, comprises at least one of:

setting a weight for each personal attribute of the variable data;

supplementing an incomplete value with a default value in the variable data;

determining a selection of repeated variable data; processing interaction effects between the variable data; or generating new variable data based on the variable data.

8. The non-transitory, computer-readable medium of claim 6 , wherein at least one of the at least two candidate insurance risk prediction models is a trained generalized linear model.

9. The non-transitory, computer-readable medium of claim 6 , wherein the GBDT uses a gamma regression objective function as an objective function, and wherein the at least one modeling target value using the GBDT includes a difference between an actual modeling target value and a predicted modeling target value.

10. The non-transitory, computer-readable medium of claim 6 , wherein the training objective function is optimized using a Stochastic Gradient Descent.

11. A computer-implemented system that identifies an insurance risk prediction model that generates a more accurate risk prediction result for a respective insurance service risk assessment and that requires personal attribute information, the computer-implemented system comprising:

a prediction server including one or more computers; and

one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, perform operations comprising:

receiving, at the prediction server, an insurance service scenario comprising variable data of personal attribute information associated with at least one insurance user;

generating, at the prediction server, at least two candidate insurance risk prediction models comprising a first model and a second model, wherein the first model comprises a neural network, and wherein the second model is constructed by modeling at least one modeling target value using a gradient boosting decision tree (GBDT), and wherein the at least one modeling target value includes at least one of a loss ratio, a claim frequency, and a claim amount of the at least one insurance user;

training, at the prediction server, the first model to include a learned continuous feature vector using a training objective function, wherein the training of the first model comprises:

randomly initializing, at the prediction server, parameters of the first model,

generating, at the prediction server, N bins each comprising data of a different feature type included in the variable data of personal attribute information,

generating, at the prediction server, a first discrete feature vector comprising at least N elements corresponding to the N bins,

mapping, at the prediction server, the first discrete feature vector and at least a second discrete feature vector to generate a candidate continuous feature vector,

inputting, at the prediction server and into the first model, the candidate continuous feature vector representing the variable data of personal attribute information,

obtaining, at the prediction server, output of the first model indicating a risk score based on the candidate continuous feature vector, and

optimizing, at the prediction server, the training objective function associated with the first model based on the risk score, the parameters of the first model, and the at least one modeling target value;

generating, at the prediction server, a first function output of a Tweedie regression objective function of the GBDT using the variable data of personal attribute information as an input to the second model;

generating, at the prediction server, a second function output of the first model using the variable data of personal attribute information that was used as an input to the first model;

generating, at the prediction server, a first statistical indicator based on the first function output and the input to the second model;

generating, at the prediction server, a second statistical indicator based on the second function output and the input to the first model;

selecting, at the prediction server and for the insurance service scenario, the insurance risk prediction model that is more statistically significant from the at least two candidate insurance risk prediction models based on a comparison of the first statistical indicator and the second statistical indicator; and

generating, at the prediction server and for the insurance service scenario, an insurance risk prediction result of the at least one insurance user using the selected insurance risk prediction model.

12. The computer-implemented system of claim 11 , wherein the operations further comprise preprocessing the variable data of personal attribute information, wherein the preprocessing, at the prediction server, comprises at least one of:

setting a weight for each personal attribute of the variable data;

supplementing an incomplete value with a default value in the variable data;

determining a selection of repeated variable data; processing interaction effects between the variable data; or generating new variable data based on the variable data.

13. The computer-implemented system of claim 11 , wherein at least one of the at least two candidate insurance risk prediction models is a trained generalized linear model.

14. The computer-implemented system of claim 11 , wherein the GBDT uses a gamma regression objective function as an objective function, and wherein the at least one modeling target value using the GBDT includes a difference between an actual modeling target value and a predicted modeling target value.

15. The computer-implemented system of claim 11 , wherein the training objective function is optimized using a Stochastic Gradient Descent.