Dynamic Baseline Alerts Now Automatically Find the Best Algorithm for You
Blog post from New Relic
New Relic has enhanced its Dynamic Baseline Alerts by implementing an auto-discovered seasonality system and an unsupervised ensemble algorithm to improve prediction accuracy for their clients. This development allows New Relic to automatically determine the seasonality of time-series data using Fast Fourier Transforms and selects the best-fitting algorithm from a set of options based on recency, trend, and seasonality factors. The ensemble system evaluates multiple algorithms, such as single and triple exponential smoothing, to find the optimal one using the MASE statistical method, often favoring simpler algorithms for data with minimal seasonality. This approach enables New Relic to manage over a billion events and metrics per minute for more than 15,000 customers, continually refining their systems to deliver real-time enhancements across their SaaS platform. Nadya Duke Boone, VP of Product Management at New Relic, emphasizes the company's commitment to leveraging data science and AI to support their customers' system management needs.