Continuous anomaly detection in satellite image time series based on Z-scores of Season-Trend model Residuals

Energy Management in an Agile Workspace using AI-driven Forecasting and Anomaly Detection

“…Since the collected data falls under the category of stationary time series, non-regressive approaches [35] such as z-score can be used [36]. This method has been utilised in detection of anomalies in many applications such as [37], [38], [39]. Z-score can be calculated as:…”

A Forecasting Approach to Online Change Detection in Land Cover Time Series

“…As the model is re-estimated at each time step this does not require any assumptions of stationarity. Furthermore it does not impose assumptions on the structure of the data as in [3], [7], and [8].…”

“…Classical change detection methods that assume stationarity can then be applied [10]. Some approaches assume the original time series is cyclo-stationary [6] while others make assumptions regarding the structure of the trend and periodicity [3], [7], [8]. These types of methods can be further split into supervised methods, which require a labeled training set to estimate class specific statistics [6], [9], and unsupervised methods, which have no such requirement [3], [5], [7], [8].…”

“…The work presented in [8] applies a parametric model identical to that of [3]. In this case, residuals from the model over the entire series are converted to z-scores and this time series is directly monitored for outliers.…”