Abstract: The present disclosure generally relates to providing accurate and real-time insights into abnormal entities and activities using machine learning algorithms. An exemplary computer-enabled method comprises receiving a set of input data, wherein the set of input data is associated with an entity; automatically obtaining, based on the received set of input data, a set of derived data, wherein the set of derived data is associated with the entity; obtaining, based on the set of derived data, a plurality of feature values corresponding to a plurality of features; providing the plurality of feature values to an autoencoder-decoder to obtain a plurality of feature-specific reconstruction errors; selecting, based on the plurality of feature-specific reconstruction errors, one or more features from the plurality of features; outputting the selected one or more features and one or more textual descriptions associated with the selected one or more features.