| Title |
Comparative analysis of similarity measures for multidimensional streaming data / |
| Translation of Title |
Daugiamačių srautinių duomenų panašumo matai. |
| Authors |
Ponomarenko, Maksym |
| Full Text |
|
| Pages |
55 |
| Keywords [eng] |
multivariate time series, similarity measure, streaming data, deep learning, autoencoders, similarity measure, biosensor, anomaly detection, wearable sensors. |
| Abstract [eng] |
This work is focused on the analysis of multivariate time series using a similarity measure in various fields. The aim of this work is to provide а robust framework for multidimensional signal analysis. The purpose of this framework should be cardiac anomaly detection via ECG (Electrocardiogram) using PPG (Photoplethysmogram) as an additional source of information via multidimensional streaming time series anomaly detection algorithms and deep learning algorithms. Some limitations have been made in order to make this framework applicable to modern tasks and implementable for wearable devices, such as computational complexity and choice of physiological signals that is realistic to acquire using wearable sensors. The key mathematical algorithms of this framework were implemented. The author managed to create a framework that is illustrated in Figure 25. This solution uses the combination of three methods: Data mining to extract cardiac cycle-like segments, Similarity similarity measure model for pre-filtering on extracted segments and Convolutional Autoencoder in order to perform more advanced anomaly detection on data that has been received after the previous model. The combination of these techniques allows to increase both detection speed, accuracy and flexibility in terms of regulatable abnormality measure threshold that can be adjusted . |
| Dissertation Institution |
Vilniaus universitetas. |
| Type |
Master thesis |
| Language |
English |
| Publication date |
2021 |
