Member-only story
Anomaly Detection with Python Library PyOD — Time Series Voltage Data Outlier Detection Example
Anomaly Detection, also known as Outlier Detection, has a wide range of applications, such as fraud detection. Those outliers are data points that greatly deviate from the other data points. As we don’t know the classification standards, detecting outliers automatically often utilizes unsupervised learning. This post briefly summarises the common detection methods and gives a simple example of detecting abnormal voltage points with PyOD autoencoder.
Common anomaly detection techniques
- Classification-based or clustering-based methods
It is intuitive to group the dataset, and those outliers with great differences will stand out by themselves. Algorithms like KNN, K-means and Isolation Forest are of this kind.
2. Statistical-based methods
For these methods, outliers can be identified by creating a statistical distribution model. For example, if a dataset follows the normal distribution, the anomaly points fall into the end parts of distribution tails.
3. Graphical-based methods
Sometimes, we could easily spot the outliers by visualizing the data points.
