Anomaly Detection for Univariate Time Series with Statistics and Deep Learning
Jian-Bin Kao, Jehn‐Ruey Jiang · 2019 IEEE Eurasia Conference on IOT, Communication and Engineering (ECICE) · 2019
This paper proposes an anomaly detection framework for univariate time series data. Based on the Dickey-Fuller test, fast Fourier transform (FFT), and Pearson product-moment correlation coefficient, data are classified into three classes, namely (1) stationary, (2) periodic and (3) non-stationary and non-periodic time series. Different schemes using statistics and gated recurrent unit (GRU) deep learning concepts are applied to different class of time series for performing anomaly detection. The proposed framework outperforms related methods, namely STL, SARIMA, LSTM, LSTM with STL, and ADSaS, in almost all measurements for five Numenta Anomaly Benchmark(NAB) datasets.