Types of Concept Drift
What you'll learn: How to distinguish between four major patterns of concept drift—sudden, gradual, incremental, and recurring—and recognize their real-world signatures.
Understanding the Four Drift Patterns
While the previous lesson introduced what concept drift is (changes in the relationship between inputs and outputs), this lesson focuses on how that change manifests over time.
Sudden (Abrupt) Drift
The relationship changes instantaneously at a specific point. Think of a regulatory change that immediately alters customer behavior, or a system update that changes how data is collected.
Real-world example: A payment fraud model breaks when a new payment method launches overnight, creating entirely new fraud patterns.
Gradual Drift
The old and new concepts coexist for a period, with the new concept slowly becoming dominant. Picture a transition phase where both old and new patterns appear in the data stream simultaneously.
Real-world example: A gradual shift in customer preferences as a new competitor enters the market—some users switch immediately, others slowly migrate over months.
Incremental Drift
The relationship changes continuously in small steps, like a slow drift rather than discrete states. Each change is tiny, but cumulative effect is significant.
Real-world example: Inflation gradually changing price sensitivity in a recommendation model, or seasonal temperature shifts affecting energy consumption patterns.
Recurring (Seasonal) Drift
Concepts repeat cyclically—the relationship alternates between known states. Unlike other drifts, the "new" concept has appeared before.
Real-world example: E-commerce models face recurring drift during holiday seasons, back-to-school periods, and sales events—patterns return predictably.
Why This Distinction Matters
Each drift type requires different detection sensitivity and retraining strategies. Sudden drift needs immediate response; incremental drift requires trend analysis; recurring drift benefits from context-aware models that "remember" seasonal patterns.
Key Takeaway: The four drift patterns—sudden, gradual, incremental, and recurring—require different detection windows and response strategies, making pattern recognition critical for effective model maintenance.