Feature Extraction of Hidden Oscillation in ECG Data via Multiple-FOD Method

Abstract

Fourier transform (FT) is a non-parametric method which can be used to convert the time domain data into the frequency domain and can be used to find the periodicity of oscillations in time series datasets. In order to detect periodic-like outliers in time series data, a novel and promising method, named as the outlier detection via Fourier transform (FOD), has been developed. From our previous studies, it has been shown that FOD outperforms most of the commonly used approaches for the detection of outliers when the outliers have periodicity with low fold changes or high sample sizes. Recently, the multiple oscillation and hidden periodic-like pattern cases for time series data have been investigated and found that the multiple application of FOD, shortly multiple-FOD, can also be a successful method in the detection of such patterns. These empirical results are based on real electrocardiogram (ECG) data where the discrimination of disorders can be helpful for the diagnosis of certain heart diseases in advance. Hereby, in this study, we evaluate the performance of multiple-FOD in different types of simulated datasets which have distinct sample sizes, percentage of outliers and distinct hidden patterns.