Mining Approximate Temporal Functional Dependencies Based on Pure Temporal Grouping

Abstract

Functional dependencies (FD) model constraints over databases like "employees with the same role get the same salary". Some extensions have been introduced to represent temporal constraints: temporal functional dependencies (TFD) represent constraints like "for any given month, employees with the same role have the same salary, but their salary may change from one month to the next one"'; approximate functional dependencies (AFD) hold on most of the facts stored by the database, enabling data to deviate from the defined constraints according to a user-defined percentage like "employees with the same role generally have the same salary". By this paper, we merge the concepts of temporal functional dependency and of approximate functional dependency, introducing the concept of approximate temporal functional dependency (ATFD). ATFD can be defined and measured either on temporal granules (e.g., grouping data by day, week, month, year) or on sliding windows (e.g., a fixed-length time interval which moves over the time axis). We also introduce some specific data mining techniques for ATFDs. As a proof of concept, we developed a running prototype, proving the feasibility of our proposal and testing it on a real-world database from the medical domain of psychiatry.