Fintech and Enterprise platforms ingest massive volumes of timestamped data (big data) from IoT devices such as payment terminals, wearables, and mobile apps. Accurate timing is essential for fraud detection, risk scoring, and customer analytics. Yet a subtle irregularity called the leap second can corrupt timestamps and trigger AI drift, gradually degrading model performance in production.
In this article, I will attempt to explain clearly what drift types are and how they can be prevented, based on my research paper. Details can be found here. Let’s start.