Table 2.
Comparison of CPD algorithm scalability based on sliding window size n. * = estimate based on algorithm.
| Category | Method | Parametric/Non Parametric | Computational Cost |
|---|---|---|---|
| Probability Density Ratio | CUSUM | Parametric | O(n2)* |
| AR | Parametric | O(n3)* | |
| KLIEP | Non Parametric | KLIEP< CUSUM ; KLIEP < AR | |
| uLSIF | Non Parametric | uLSIF < KLIEP | |
| RuLSIF | Non Parametric | RuLSIF < uLSIF | |
| SPLL | Semi Parametric | O(n2)* | |
| Subspace Models | SI | Parametric | SI > KLIEP |
| SST | Parametric | SST > KLIEP | |
| Probabilistic Method | Bayesian | Parametric | O(n) |
| GP | Non Parametric | O(n2) | |
| Kernel Based Methods | KcpA | Non Parametric | O(n3) |
| Clustering | SWAB | O(Ln) | |
| MDL | |||
| Shapelet | |||
| Model Fitting | |||
| Graph Based Methods | Non Parametric | ||
| Multi-Class Classifier | Nearest Neighbor | Non Parametric | = Cost (Training + CP detection) |
| HMM | Parametric | ||
| GMM | Parametric | ||
| Binary Class Classifier | SVM | Parametric | |
| Naive Bayes | Parametric | ||
| Logistic Regression | Parametric |