Assessing accuracy degradation over time with a Markov-chain model

Accuracy, among the most discussed data quality dimensions in literature, reflects the extent to which data values match a baseline perceived to be correct – e.g., the true real-world attribute values, or another validated dataset. Even when data values are accurate when acquired, their accuracy may degrade over time - certain properties of real-world entities may change, while the data values that reflect them are not being updated. Drawing on that assumption, this study suggests a Markov-Chain model that describes accuracy degradation over time – this by assessing the likelihood of a data attribute to transition from one state to another within a given time period. Evaluation of the model with real-world data shows its potential contribution for a few key data-quality management tasks, such as the prediction of accuracy degradation, and the development of data auditing and maintenance policies.