A non-transitory, computer-readable storage medium storing computer-executable instructions that, when executed with one or more processors of a computing device, causes the one or more processors to:
- generate, from legitimate account data instances using byte pair encoding (BPE), a data compression vocabulary, the data compression vocabulary developed with the BPE and specifying a plurality of tokens identified from the legitimate account data instances, each token of the plurality of tokens being individually associated with a corresponding replacement value, wherein the BPE is tailored to a format and structure of the account data;
- during and as at least part of an online account creation process:
- receive, via a network, an account creation request to generate a user account, the account creation request comprising account data having an initial length measured in bytes or characters;
- and (ii) comparing the compression ratio metric to a predefined threshold, the predefined threshold value being determined based on historical compression ratios of legitimate account data and dynamically adjusted in real-time based on updated compression ratios of newly validated legitimate account data;
- applying, by the computing device, the data compression vocabulary to compress the account data and generate compressed account data based at least in part on replacing a token identified in the account data with a respective replacement value from the data compression vocabulary;
- determine through automated compression analysis whether the account data is fraudulent based at least in part on:
- (i) calculating a compression ratio metric and a predefined threshold value, wherein the compression ratio metric is based at least in part on a difference between the initial length of the account data and a resultant length of the compressed account data measured in bytes or characters;
- in response to determining that the account creation request is fraudulent according to the automated compression analysis, execute a first set of operations to:
- automatically deny the account creation request;
- forgo execution of operations associated with generating the user account;
- and
- log the account creation request as a potential fraud incident in a fraud detection database, wherein the account creation request is determined to be fraudulent based on determining that the resultant length of the compressed account data as compared to the initial length of the account data fails to exceed the predefined threshold value;
- and
- in response to determining that the account creation request is legitimate according to the automated compression analysis, execute a second set of operations for:
- generating the user account using the account data;
- updating the data compression vocabulary using the account data, wherein the updating comprises:
- identifying new tokens in the account data not present in the data compression vocabulary;
- adding the new tokens to the data compression vocabulary with replacement values derived from the BPE;
- dynamically adjusting the predefined threshold value based on updated compression ratios of legitimate account data;
- and
- storing the updated data compression vocabulary in a data store.