# Data Quality

Data quality is a contingent concept - exactly what constitutes quality will be highly dependent on the audience for the data. Further, research shows that '[strange things' such as errors and outliers in the data may actually help users ](https://user-centric-open-data-publishin.gitbook.io/user-centric-open-data-publishing/preparation/data-quality-2)engage with the data by providing an opportunity for them to question the data. 

The first step in the data quality process is map out exactly what your version of data quality looks like.

<figure><img src="https://2677491958-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F4tiWPkv75at4eCXxyJ7O%2Fuploads%2FEO4rgThZf5lG73KND1Cd%2FML%20quality%20planning.png?alt=media&#x26;token=6a960b60-dbdb-4905-a164-f37cfab7962a" alt=""><figcaption><p>The 4 steps of data quality </p></figcaption></figure>

Source: [A Survey of Data Quality Requirements That Matter in Machine Learning Pipelines](https://dl.acm.org/doi/10.1145/3592616)