Hi Armin,

We're happy that you found the blog post interesting!

Previously unseen events are notoriously difficult to handle for ML algorithms. In FreqAI, we have implemented a series of outlier identification methods that help us detect such events (you can read more about them here: https://www.freqtrade.io/en/latest/freqai-feature-engineering/#outlier-detection). You as a user can then decide how to use that information: Do you want to enter/exit trades during such events or should your bot disengage until the market has stabilized or your model has seen enough such events to be able to properly characterize them?

We are prepping a new blog post describing one of our experiments with FreqAI where we used the Dissimilarity Index (https://www.freqtrade.io/en/latest/freqai-feature-engineering/#identifying-outliers-with-the-dissimilarity-index-di) as part of our trading strategy so be on the look-out for that to see a real-life use-case!