Risky transaction identification method and apparatus
A feature extraction is performed on transaction data to obtain a user classification feature and a transaction classification feature. A first dimension feature is constructed based on the user classification feature and the transaction classification feature. A dimension reduction processing is performed on the first dimension feature to obtain a second dimension feature. A probability that the transaction data relates to a risky transaction is determined based on a decision classification of the second dimension feature, where the decision classification is based on a pre-trained deep forest network including a plurality of levels of decision tree forest sets.
1. A computer-implemented method, comprising:
training a deep forest model that includes multiple levels of respective base classifiers on training data that specifies a plurality of transaction samples, comprising:
collecting a plurality of black samples and white samples, wherein each black sample relates to a risky transaction, and wherein each white sample relates to a normal transaction;
extracting feature data from data associated with the black samples and data associated with the white samples;
generating sampled feature data from the feature data; and
iteratively performing a training process on the deep forest model, wherein the training process comprises, for a current level of respective base classifiers:
training each base classifier included in the current level on the sampled feature data;
concatenating one or more output features of the current level to features from the sampled feature data to generate concatenated features;
training each base classifier included in a next level by using the concatenated features; and
terminating the training process upon determining that a predetermined termination condition is satisfied;
after the training, obtaining new feature data describing a transaction initiated by a user of a transaction service, wherein the new feature data comprises a set of features belonging to respective feature categories;
for each feature category:
determining, based at least on a sampling density used in collecting the features belonging to the feature category, a respective rate for use in selecting sampled features; and
selecting, from the features belonging to the feature category and in accordance with the respective rate, a plurality of sampled features; and
generating, based on processing the sampled features using the deep forest model, an output specifying a predicted classification of the transaction.
2. The computer-implemented method of claim 1 , wherein generating the sampled feature data from the feature data comprises:
performing a dimension reduction process on the feature data having a first dimension to obtain the sampled feature data that has a lower dimension than the first dimension.
3. The computer-implemented method of claim 1 , 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 an indicator that evaluates a performance of the base classifier.
4. The computer-implemented method of claim 1 , 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.
5. A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:
training a deep forest model that includes multiple levels of respective base classifiers on training data that specifies a plurality of transaction samples, comprising:
collecting a plurality of black samples and white samples, wherein each black sample relates to a risky transaction, and wherein each white sample relates to a normal transaction;
extracting feature data from data associated with the black samples and data associated with the white samples;
generating sampled feature data from the feature data; and
iteratively performing a training process on the deep forest model, wherein the training process comprises, for a current level of respective base classifiers:
training each base classifier included in the current level on the sampled feature data;
concatenating one or more output features of the current level to features from the sampled feature data to generate concatenated features;
training each base classifier included in a next level by using the concatenated features; and
terminating the training process upon determining that a predetermined termination condition is satisfied;
after the training, obtaining new feature data describing a transaction initiated by a user of a transaction service, wherein the new feature data comprises a set of features belonging to respective feature categories;
for each feature category:
determining, based at least on a sampling density used in collecting the features belonging to the feature category, a respective rate for use in selecting sampled features; and
selecting, from the features belonging to the feature category and in accordance with the respective rate, a plurality of sampled features; and
generating, based on processing the sampled features using the deep forest model, an output specifying a predicted classification of the transaction.
6. The non-transitory, computer-readable medium of claim 5 , wherein generating the sampled feature data from the feature data comprises:
performing a dimension reduction process on the feature data having a first dimension to obtain the sampled feature data that has a lower dimension than the first dimension.
7. A computer-implemented system, 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:
training a deep forest model that includes multiple levels of respective base classifiers on training data that specifies a plurality of transaction samples, comprising:
collecting a plurality of black samples and white samples, wherein each black sample relates to a risky transaction, and wherein each white sample relates to a normal transaction;
extracting feature data from data associated with the black samples and data associated with the white samples;
generating sampled feature data from the feature data; and
iteratively performing a training process on the deep forest model, wherein the training process comprises, for a current level of respective base classifiers:
training each base classifier included in the current level on the sampled feature data;
concatenating one or more output features of the current level to features from the sampled feature data to generate concatenated features;
training each base classifier included in a next level by using the concatenated features; and
terminating the training process upon determining that a predetermined termination condition is satisfied;
after the training, obtaining new feature data describing a transaction initiated by a user of a transaction service, wherein the new feature data comprises a set of features belonging to respective feature categories;
for each feature category:
determining, based at least on a sampling density used in collecting the features belonging to the feature category, a respective rate for use in selecting sampled features; and
selecting, from the features belonging to the feature category and in accordance with the respective rate, a plurality of sampled features; and
generating, based on processing the sampled features using the deep forest model, an output specifying a predicted classification of the transaction.
8. The computer-implemented system of claim 7 , wherein generating the sampled feature data from the feature data comprises:
performing a dimension reduction process on the feature data having a first dimension to obtain the sampled feature data that has a lower dimension than the first dimension.
9. The computer-implemented method of claim 1 , wherein
selecting, from the features belonging to the feature category and in accordance with the respective rate, a plurality of sampled features comprises:
performing feature sampling on the features belonging to the feature category in accordance with the respective rate.
10. The computer-implemented method of claim 1 , wherein the predicted classification of the transaction comprises a risky classification or a normal classification.