Validation of account identifier

A system for training a model to predict whether an account action will fail due to an invalid account identifier, the system comprising:

• one or more processors;
• and
• at least one memory coupled to the one or more processors and storing instructions that, when executed by the one or more processors, cause the system to perform operations including:
• retrieving historical data including a plurality of character sequences each entered for identifying an account to be used in the account action, each character sequence including at least a first portion;
• grouping the character sequences based on the first portions;
• extracting a set of trimmed subsequences from each respective group of character sequences, each trimmed subsequence having a number of trimmed characters;
• grouping ones of the trimmed subsequences based on the numbers of trimmed characters;
• determining, for each first portion, for each group of trimmed subsequences, a ratio of a number of unique subsequences included in the group of trimmed subsequences to a total number of instances the corresponding first portion is used in the historical data;
• determining, for each respective character sequence and each subsequence extracted from the respective character sequence, a feature value for each of a set of features defined for characterizing sequences, the determining based on the historical data, a number of characters included in at least one portion of the respective character sequence, the determined ratios, and sequences of characters submitted during one or more recent time periods during which the detected character sequence was submitted;
• and
• training the model, using model training data in conjunction with a machine learning (ML) algorithm, to generate a predictive value suggestive of a likelihood that a given sequence, if submitted, will result in an account action failure, the model training data generated based on the feature values and a set of labels assigned to the plurality of character sequences, and the predictive value generated based at least in part on rates of success associated with the one or more recent time periods, wherein one or more actions for preempting submission of the given sequence are selectively performed based on whether the generated predictive value is greater than a threshold.

A method for using a model trained in conjunction with a machine learning (ML) algorithm to predict whether an account action will fail due to an invalid account identifier, the method performed by one or more processors of a validation system and comprising:

• detecting a character sequence being entered for identifying an account to be used in the account action, the character sequence including at least a first portion;
• grouping the character sequence based on the first portion;
• extracting a set of trimmed subsequences from the group of character sequences, each trimmed subsequence having a number of trimmed characters;
• grouping ones of the trimmed subsequences based on the numbers of trimmed characters;
• determining, for the first portion, for each group of trimmed subsequences, a ratio of a number of unique subsequences included in the group of trimmed subsequences to a total number of instances the first portion is used in a set of historical data;
• determining, for the character sequence and each trimmed subsequence extracted from the character sequence, a feature value for each of a predefined set of features for characterizing sequences, the determining based on a number of characters included in at least one portion of the character sequence, the determined ratios, and sequences of characters submitted during one or more recent time periods during which the detected character sequence was submitted;
• generating, using the feature values as input to the model, a predictive value suggestive of a likelihood that the detected character sequence, if submitted, will result in the account action failing, the predictive value generated based at least in part on rates of success associated with the one or more recent time periods;
• and
• selectively performing a preemptive action for preempting submission of the character sequence based on whether the predictive value is greater than a threshold.