Raw data usually contains a lot of discrepancies in the form of formatting issues, missing data/labels to name a few. Broadly speaking, data could be incomplete, inconsistent and noisy.
It is imperative to rectify such issues before visualizing or analyzing it to get meaningful results. This is also called as Data Cleaning. However, Data Preprocessing is a much broad umbrella under which many operations are done in order to get the data in the required state. 
Checking the descriptive stats of the data is one of the initial steps to see for discrepancies. Missing value imputations, encoding categorical features, reducing Multicollinearity, scaling, etc. are the first few steps.
However, to fully prepare the data for modelling, much more preprocessing needs to be done in order to give the best possible set of features to the model.
Data Preprocessing is different for different types of data. Unstructured data like text need to be cleaned and useful information is needed to be extracted by Tokenization, Lemmatization, etc. Image data needs to resized/reshaped, augmented, etc.
This is just the tip of the ice. Data Preprocessing includes a lot more custom steps as well which depend on the type of data or the problem.