Change Detection with Kalman Filter and CUSUM

摘要: In most challenging applications learning algorithms acts in dynamic environments where the data is collected over time. A desirable property of these ability incremental incorporating new actual decision model. Several have been proposed. However them make assumption that examples are drawn from a stationary distribution [13]. The aim this study to present detection system (DSKC) for regression problems. modular and works as post-processor regressor. It composed by predictor, Kalman filter Cumulative Sum Recursive Residual (CUSUM) change detector. continuously monitors error significant increase interpreted generates When detected, model deleted one constructed. paper we tested DSKC with set three artificial experiments, two real-world datasets: Physiological dataset clinic Sleep Apnoea. Apnoea common disorder characterized periods breathing cessation (apnoea) reduced (hypopnea) [7]. This real-application goal detect changes signals monitor breathing. experimental results showed detected fast high probability. also robust false alarms can be applied efficiency problems information available