Automatic Discovery of Functional Dependencies and Conditional Functional Dependencies: A Comparative Study

Over the last twenty years, several algorithms have been proposed for automatic rule/constraint discovery from data, for the purpose of data cleaning. These algorithms look for constraints such as functional dependencies (FDs), conditional FDs (CFDs), inclusion dependencies (INDs), conditional INDs (CINDs), association rules, integrity constraints (ICs) and denial constraints (DCs), among others. While some of these techniques are direct generalizations and extensions of others, many differ greatly from the rest in approach, characteristics and general flavour. Many of these algorithms have not been tested and compared against each other, therefore their core differences and relative strengths and weaknesses are hard to comprehend.