Raw data are data that have not been processed for meaningful use. A raw dataset is exactly what is collected, before any data cleaning, processing or analysis has been completed. 

It is often useful to store raw data as well as the cleaned, processed data, as it can help your work to be more easily reproduced. If another researcher has your raw data and the steps you used to process and analyse, they can recreate your results. This has to be balanced with the cost of storing raw data, and the likelihood of the raw data being useful compared to data that has undergone an initial process of data cleaning.