What are three key monitoring metrics for deployed CPMAI models?

• A. Data input distribution drift; system health; training time
• B. Prediction accuracy drift; data input distribution drift; system health
• C. Model size; training dataset size; UI color scheme
• D. Server location; time of day; dataset age

Answer: (B)

Monitoring deployed CPMAI models focuses on three things: how well the model keeps making correct predictions, whether the inputs it sees have shifted, and whether the service itself remains reliable. The best trio to watch is prediction accuracy drift, data input distribution drift, and system health.

Prediction accuracy drift tracks changes in how often the model’s predictions are correct over time on live data. If accuracy starts to fall, the model may be misaligned with current patterns and might need retraining or adaptation.

Data input distribution drift looks at shifts in the input features compared with what the model was trained on. When the input data changes—different ranges, new features, or different patterns—the model’s performance can degrade, even if its internal logic hasn’t changed. Detecting drift helps you decide when to refresh data pipelines or retrain.

System health covers the operational side: inference latency, throughput, error rates, uptime, and resource usage. Even a highly accurate model can deliver poor user experiences if the system it runs on is slow or unstable.

The other options mix in elements that aren’t standard ongoing production monitoring metrics—for example, training time is a development-stage concern, and items like UI color schemes, server location, or dataset age aren’t core indicators of how a deployed model is performing or serving users.