This approach considers the most recent time series with the size of a specific window and trains the model. Then it classifies the data as follows:
- Takes the most recent time series with window size used for training
- Classifies it—which of the clusters does it belong to?
- Uses the ML model for that cluster to predict the price or price change
This solution dates back to 2014, but still it gives a certain level of robustness. By having many parameters to identify, and not having the order-book historical data available easily, in this project, we use a simpler approach and dataset.
