Feature Selection Risks

Feature selection risks involve choosing the wrong input variables for a machine learning model, which can lead to spurious correlations. In finance, there are thousands of potential features, from technical indicators to on-chain metrics and social media sentiment.

If a model is allowed to select features without a strong economic hypothesis, it may find patterns that are purely coincidental. For example, a model might correlate the price of a crypto asset with an unrelated global event, leading to poor performance when that correlation inevitably breaks down.

In derivatives trading, features must be relevant to the underlying price dynamics or volatility structures. Selecting features based on economic theory rather than purely statistical significance helps mitigate this risk.