Deep learning based anomaly detection in time-series data
Anomaly detection, also called outlier detection, on the multivariate time-series data is applicable to multiple domains. With the proliferation of deep learning-based methods, we aim to leverage on them to tackle anomaly detection, mainly on the field of industry data (server machines and spacecraf...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2020
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/137949 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Anomaly detection, also called outlier detection, on the multivariate time-series data is applicable to multiple domains. With the proliferation of deep learning-based methods, we aim to leverage on them to tackle anomaly detection, mainly on the field of industry data (server machines and spacecrafts, which are monitored with multivariate time series). This project proposes an anomaly detection framework, which includes data exploration, data pre-processing, RNN-based models, dynamic threshold selection. The effectiveness of various machine learning technologies such as long short-term memory networks (LSTMs), gated recurrent unit networks (GRUs) and autoencoders (AEs) are examined. Subsequently, for dynamic threshold selection, a non-parametric and computationally efficient approach is proposed. The error threshold pruning is introduced to mitigate false positives. The final result demonstrates the capability of the proposed framework for anomaly detection on the multivariate time-series data in the industry data. |
|---|