A new frequency domain and dynamic time warping based feature: WFOD feature

Anomaly detection is one of the critical steps in the diagnosis of disorders. Many anomaly detection methods aim to use different characteristics of the anomalous data with distinctive features than the normal data. We propose a new feature extraction method, the WFOD feature which is based on a novel data transformation in the frequency domain called WFOD. In our analyses, we use a publicly available motion artifact ECG dataset, which are collected in three different cases: standing, walking and jumping. Such cases are classified by four different classifiers with the pairs of statistical moments of the data, with and without the WFOD feature. The results show that the WFOD feature enhances the classification accuracy in most cases by improving accuracy by up to 25% on accuracy values ranging from the worst case, 47% to the best, 93%