Abstract: Concrete data types for raw data organized in one or more columns of one or more tables may be determined. Functional data types for the one or more columns may be determined based on the raw data and the concrete data types such that a portion of the one or more columns may be associated with an identifier data type. Relationships between the one or more tables may be determined based on the portion of the one or more columns associated with the identifier data type. A schema representing the raw data may be generated based on the one or more relationships and the one or more tables.

Abstract: Embodiments are directed to managing machine learning models. Domain items may be determined based on domain data and schema information. Labels that correspond to a predicted outcome may be generated based on the domain data. A model may be trained based on a portion of a plurality of feature records and the labels such that each feature record may be associated with an observance of a domain item. The trained model may be disqualified based on evaluation metrics that may be below a threshold value causing further actions, including: submitting other portions of the feature records to the disqualified model; determining erroneous feature fields in the feature records based on metrics associated with the submission of the other portions of feature records; updating the feature records to exclude the erroneous feature fields; retraining the disqualified model based on the updated feature records.