Grupo Alarcos · Journal article · 2026
DMN4DQ+: Optimising data repair to enhance data usability
Álvaro Valencia-Parra, Ángel Jesús Varela-Vaca, Luisa Parody, Ismael Caballero Muñoz-Reja, Maria Teresa Gomez-Lopez
• We enable the characterisation of a dataset’s usability by introducing the concept of a usability profile. • We propose a framework for modelling corrective actions to address the various data quality problems, enabling users to define the requirements and target usability. • We propose an approach based on the Constraint Optimisation paradigm to select the optimal set of corrective actions to apply to enhance the usability of data. • The DMN4DQ+ methodology includes a data quality life cycle formed of data quality rule descriptions, usability assessments, and corrective action selections for data repair. Data quality has become crucial in decision-making and data analysis. There is an intrinsic relationship between data quality and usability; however, acceptable levels of data quality depend on the contextual requirements and operational priorities of an organisation. The context of use, business needs, and the organisation’s appetite for risk all influence the usability of data. Achieving adequate levels of usability sometimes requires specific corrections, which can be costly and may incur have far-reaching consequences. This paper introduces the concept of target usability, by representing the minimum level of usability determined by business analysts at which data records can be used without compromising organisational performance. When data records fail to meet this threshold, a combination of corrective actions can be implemented to improve both quality and usability. To achieve near-optimal outcomes, the data quality analyst can effectively combine these sets of actions. This paper proposes DMN4DQ+, an extension of DMN4DQ, where the optimal combination of corrective actions can be derived from the application of constraint optimisation techniques based on the data quality rules described in decision models, the cost model of the actions, and on the usability profile of the record to be improved. The development of a technological stack has conveniently supported DMN4DQ+, and it has been evaluated using a real dataset, thereby demonstrating its applicability and performance.