In data science, the most important thing is data. The data holds the ground truth about any events, phenomena, or experiments that are going on around us. Once we've processed the data, we get information. Once we've processed this information, we can derive knowledge from it. Hence, the most prominent stage in knowledge extraction is how relevant the data that's being captured is. There are different types of data, including structured data, unstructured data, and semi-structured data. Structured data maintains a uniform structure in all the observations, similar to relational database tables. Unstructured data does not maintain any particular structure. Semi-structured data maintains some structure in the observation. JavaScript Object Notation (JSON) is one of the most popular ways to store semi-structured data. 

The process of collecting data in any company depends on the kind of project and the type of information that needs to be studied. The different types of datasets range from text data, file, database, sensors data, and many other Internet of Things (IoT) data. However, when learning about a machine learning workflow, most students prefer to avoid the data collection phase and use open source data from places such as Kaggle and the UCI Machine Learning Repository.