Review:
One Class Svm
overall review score: 4.2
⭐⭐⭐⭐⭐
score is between 0 and 5
The One-Class SVM (Support Vector Machine) is an unsupervised learning algorithm used primarily for novelty detection and outlier detection. It learns a decision boundary around the majority of data points in a high-dimensional feature space, effectively modeling the distribution of a single class to identify whether new instances belong to that class or are deviations/outliers.
Key Features
- Unsupervised learning technique targeting novelty and outlier detection
- Builds a boundary around data points in feature space using kernel functions
- Effective in high-dimensional spaces
- Requires only data from the 'normal' class during training
- Utilizes Support Vector Machine principles to find the minimal enclosing boundary
Pros
- Highly effective for anomaly detection when only normal data is available
- Flexible with choice of kernel functions to adapt to complex data distributions
- Robust to high-dimensional data
- Widely used in fraud detection, network security, and anomaly identification
Cons
- Performance depends heavily on parameter tuning (e.g., kernel type, nu parameter)
- Sensitive to imbalanced datasets with few anomalies, which may affect accuracy
- May produce false positives if the normal class has high variability
- Requires careful selection of hyperparameters for optimal results