Avoiding drifts in your models

Data and concept drifts challenge the reliability of machine learning models in production. Drifts in our machine learning projects can have different characteristics. Some of these characteristics that could help you to detect drifts in your projects and plan to resolve them are as follows:

• Magnitude: We might face magnitudes of difference across the data distribution that result in drift in our machine learning models. Small changes in the data distribution may be difficult to detect, while large changes may be more noticeable.
• Frequency: Drifts might occur in different frequencies.
• Gradual versus sudden: Data drift can occur gradually where changes in the data distribution happen slowly over time, or it can occur suddenly where changes happen quickly and unexpectedly.
• Predictability: Some types of drift may be predictable, such as changes that occur seasonally or due to external events. Other types of drift may be unpredictable...