This paper aims at comparing two local outliers detection techniques. One is based on a Least Squares Support Vector Machine technique within a sliding window-based learning algorithm. A modification is proposed to improve its performance in non-stationary time-series. The second method relies on the Principal Component Analysis theory along with a robust orthonormal projection approximation subspace tracking with rank-1 modification. The comparative performance of these methods are assessed through simulations using a non stationary time-series generated with a nonlinear input-output model.
|Title of host publication||2016 12th IEEE International Conference on Control and Automation (ICCA)|
|Number of pages||6|
|Publication status||Published - 1 Jun 2016|
|Name||IEEE International Conference on Control and Automation ICCA|
- WIRELESS SENSOR NETWORKS
- ANOMALY DETECTION
Gil, P., Martins, H., Cardoso, A., & Palma, L. (2016). Outliers detection in non-stationary time-series: Support vector machine versus principal component analysis. In 2016 12th IEEE International Conference on Control and Automation (ICCA) (pp. 701-706). (IEEE International Conference on Control and Automation ICCA). IEEE. https://doi.org/10.1109/ICCA.2016.7505361