An approach for alert raising in real-time data warehouses

Abstract

This work proposes an approach for alert raising within a real-time data warehouse environment. It is based on the calculation of confidence intervals for measures from historical facts. As new facts arrive to the data warehouse on a real-time basis, they are systematically compared with their appropriate confidence intervals and alerts are raised when anomalies are detected. The interest of this approach is illustrated using the particular real world use case of technical analysis of stock data.