Most common is z-score, however may not always be optimal
Might be worth incorporating normalisation techniques into model tuning
Is it better to be normalising % change in mid-price rather than change itself as changes in mid-price may be extremely small and therefore lost in normalisation?
Work in Progress notes: https://dsmp.atlassian.net/wiki/x/AwDR
Some interesting points: