Risky transaction identification method and apparatus

1. A computer-implemented method of identifying a risky transaction, comprising:

obtaining transaction data describing a transaction initiated by a user of a transaction service;

extracting, from the obtained transaction data, feature data having a first dimension, the feature data comprising a plurality of user classification features and a plurality of transaction classification features;

performing a dimension reduction process on the feature data having the first dimension to obtain sampled feature data having a second dimension, comprising:

for each of the plurality of user classification features:

determining a respective first feature sampling frequency; and

sampling, from the plurality of user classification features and in accordance with the respective first feature sampling frequencies, a plurality of sampled user classification features;

for each of the plurality of transaction classification features:

determining a respective second feature sampling frequency; and

sampling, from the plurality of transaction classification features and in accordance with the respective second feature sampling frequencies, a plurality of sampled transaction classification features; and

determining a probability that the transaction data relates to a risky transaction based on a decision classification of the sampled feature data, wherein the decision classification is based on a pre-trained deep forest network including a plurality of levels of decision tree forest sets.

2. The computer-implemented method of claim 1 , wherein each level of the decision tree forest set includes a plurality of base classifiers.

3. The computer-implemented method of claim 1 , further comprising training a deep forest network on training data that includes a plurality of transaction samples to obtain the pre-trained deep forest network.

4. The computer-implemented method of claim 3 , wherein training the deep forest network comprises:

collecting a plurality of black samples and white samples, wherein the each black sample relates to a risky transaction, and wherein each white sample relates to a normal transaction;

extracting, from data associated with the black samples and data associated with the white samples, feature data;

performing a dimension reduction process on the feature data to obtain sampled feature data; and

iteratively performing a training process on the deep forest network to obtain the pre-trained deep forest network, wherein the pre-trained deep forest network includes multiple levels of decision tree forest sets.

5. The computer-implemented method of claim 4 , wherein iteratively performing a training process comprises:

training each base classifier included in a first-level decision tree forest set on the sampled feature data;

concatenating one or more output features of a current-level decision tree forest set to the features from the sampled feature data;

training each base classifier included in a next-level decision tree forest set by using the concatenated features; and

terminating the training process if it is determined that a predetermined end condition is satisfied.

6. The computer-implemented method of claim 5 , wherein a number of the black samples is not equal to a number of the white samples, and the method further comprises:

prior to training each base classifier:

dividing data associated with the black samples and data with the white samples through a k-fold cross validation into one or more training datasets and one or more corresponding validation datasets;

training a base classifier on the training datasets; and

testing the base classifier on the validation datasets to obtain a indicator that evaluates a performance of the base classifier.

7. The computer-implemented method of claim 4 , further comprising:

determining a maximum decision tree depth threshold based on a black-to-white sample ratio; and

setting a maximum value of the decision tree depth to the maximum depth threshold.

8. A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations to identify a risky transaction comprising:

obtaining transaction data describing a transaction initiated by a user of a transaction service;

extracting, from the obtained transaction data, feature data having a first dimension, the feature data comprising a plurality of user classification features and a plurality of transaction classification features;

performing a dimension reduction process on the feature data having the first dimension to obtain a sampled feature data having a second dimension, comprising:

for each of the plurality of user classification features:

determining a respective first feature sampling frequency; and

sampling, from the plurality of user classification features and in accordance with the respective first feature sampling frequencies, a plurality of sampled user classification features;

for each of the plurality of transaction classification features:

determining a respective second feature sampling frequency; and

sampling, from the plurality of transaction classification features and in accordance with the respective second feature sampling frequencies, a plurality of sampled transaction classification features; and

determining a probability that the transaction data relates to a risky transaction based on a decision classification of the sampled feature data, wherein the decision classification is based on a pre-trained deep forest network including a plurality of levels of decision tree forest sets.

9. The non-transitory, computer-readable medium of claim 8 , wherein each level of the decision tree forest set includes a plurality of base classifiers.

10. The non-transitory, computer-readable medium of claim 8 , the operations further comprising training a deep forest network on training data that includes a plurality of transaction samples to obtain the pre-trained deep forest network.

11. The non-transitory, computer-readable medium of claim 10 , wherein training the deep forest network comprises:

collecting a plurality of black samples and white samples, wherein the each black sample relates to a risky transaction, and wherein each white sample relates to a normal transaction;

extracting, from data associated with the black samples and data associated with the white samples, feature data;

performing a dimension reduction process on the feature data to obtain sampled feature data; and

iteratively performing a training process on the deep forest network to obtain the pre-trained deep forest network, wherein the pre-trained deep forest network includes multiple levels of decision tree forest sets.

12. The non-transitory, computer-readable medium of claim 11 , wherein iteratively performing a training process comprises:

training each base classifier included in a first-level decision tree forest set on the sampled feature data;

concatenating one or more output features of a current-level decision tree forest set to the features from the sampled feature data;

training each base classifier included in a next-level decision tree forest set by using the concatenated features; and

terminating the training process if it is determined that a predetermined end condition is satisfied.

13. The non-transitory, computer-readable medium of claim 11 , wherein a number of the black samples is not equal to a number of the white samples, and the operations further comprise:

prior to training each base classifier:

dividing data associated with the black samples and data with the white samples through a k-fold cross validation into one or more training datasets and one or more corresponding validation datasets;

training a base classifier on the training datasets; and

testing the base classifier on the validation datasets to obtain a indicator that evaluates a performance of the base classifier.

14. The non-transitory, computer-readable medium of claim 11 , the operations further comprising:

determining a maximum decision tree depth threshold based on a black-to-white sample ratio; and

setting a maximum value of the decision tree depth to the maximum depth threshold.

15. A computer-implemented system of identifying a risky transaction, comprising:

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 one or more operations comprising:

obtaining transaction data describing a transaction initiated by a user of a transaction service;

extracting, from the obtained transaction data, feature data having a first dimension, the feature data comprising a plurality of user classification features and a plurality of transaction classification features;

performing a dimension reduction process on the feature data having the first dimension to obtain sampled feature data having a second dimension, comprising:

for each of the plurality of user classification features:

determining a respective first feature sampling frequency; and

sampling, from the plurality of user classification features and in accordance with the respective first feature sampling frequencies, a plurality of sampled user classification features;

for each of the plurality of transaction classification features:

determining a respective second feature sampling frequency; and

sampling, from the plurality of transaction classification features and in accordance with the respective second feature sampling frequencies, a plurality of sampled transaction classification features; and

determining a probability that the transaction data relates to a risky transaction based on a decision classification of the sampled feature data, wherein the decision classification is based on a pre-trained deep forest network including a plurality of levels of decision tree forest sets.

16. The computer-implemented system of claim 15 , the operations further comprising training a deep forest network on training data that includes a plurality of transaction samples to obtain the pre-trained deep forest network.

17. The computer-implemented system of claim 16 , wherein training the deep forest network comprises:

collecting a plurality of black samples and white samples, wherein the each black sample relates to a risky transaction, and wherein each white sample relates to a normal transaction;

extracting, from data associated with the black samples and data associated with the white samples, feature data;

performing a dimension reduction process on the feature data to obtain sampled feature data; and

iteratively performing a training process on the deep forest network to obtain the pre-trained deep forest network, wherein the pre-trained deep forest network includes multiple levels of decision tree forest sets.