Minimizing False Positives in Anomaly Detection Models
We at Taboola built our own in-house anomaly detection framework, for detecting issues in critical metrics. Our framework contains hundreds of models, which sometimes become noisy. In order to minimize the number of false positives, we chose different strategies such as specific cutoffs, hyperparameter tuning and integration with internal information.
Interested? Come hear Gali Katz, Senior Software Engineer at the Infrastructure Engineering Group at Taboola
Talk language: Hebrew
For the past 4 years, I worked as a senior, full-stack developer in the Production Engineering team at Taboola. Facing real-life scaling and reliability problems, I worked hard to make our monitoring framework stable and reliable. One major step during this project was the development of an anomaly detection engine, which is now becoming widely used by different teams across the company.