Anomaly Detection in Multi-Seasonal Time Series Data

Most of today’s time series data contain anomalies and multiple seasonalities, accurate anomaly detection in these is critical to almost any type business. However, most mainstream forecasting models used for can only incorporate one or no seasonal component into their forecasts cannot capture every known pattern data. In this paper, we propose a new multi-seasonal model that extends the popular Seasonal Autoregressive Integrated Moving Average (SARIMA) model. Our model, named multi-SARIMA, utilizes dataset’s pre-determined trends increase accuracy even more than original SARIMA experimental results demonstrate higher multi-SARIMA when seasonalities are present with component, although processing time.