Model score recall for machine learning systems and applications

A computer program product comprising a non-transitory computer readable storage medium having instructions encoded thereon that, when executed by a processor, cause the processor to:

• receive a scored dataset, wherein the scored dataset is a training dataset that has been scored by a machine learning model, the scored dataset comprising scores associated with input rows of cross features value within the dataset;
• flag certain features within the scored dataset, the flagged features being identified as having low relevance to the scores provided by the machine learning model;
• convert certain feature values within the scored dataset, the converted feature values identified as feature values that have a high likelihood of receiving similar scores from the machine learning model;
• compress the received scored dataset, the compressed scored dataset comprising features that have been flagged as having low relevance to scores provided by the machine learning model, and comprising converted feature values;
• receive new scoring request;
• prepare the new scoring request;
• determine whether the prepared new scoring request matches an input row of cross feature values in the compressed scored dataset;
• retrieve score associated with a matching input row of cross feature values in the compressed scored dataset if a match is determined between the prepared new scoring request and an input row of cross feature values in the compressed scored dataset;
• pass the new scoring request to a scoring system if the prepared new scoring request does not match an input row of cross feature values in the compressed dataset;
• and
• obtain a new score provided by the scoring system for a new scoring request that does not match an input row of cross features values in the compressed dataset, and determine whether to store the new scoring request and the obtained new score as an input row of cross feature values in the compressed dataset.