Hello,

yes, that is the best practice. Doing EDA first will show you what anomalies or biases are in your data. That can then help inform you what kind of preprocessing may be needed.