|
Optimization level | Ensemble learning phase | Ensemble level | Strategy adopted | Method employed |
|
Decision optimization |
Ensemble integration |
Combination level | Fusion | Majority voting method [70–72] |
Threshold plurality vote method [73] |
Naïve Bayes method [74, 75] |
Fuzzy theory method [76, 77] |
Decision template method [78] |
Metalearning method [79] |
Hierarchically structured method [82, 83] |
Boolean combination method [2] |
Selection | The test and select method [71] |
Cascading classifiers method [85] |
Dynamic classifier selection method [86, 87] |
Clustering-based selection method [17, 45, 88, 91] |
Statistical selection method [89] |
Mixture of expert systems | Stochastic selection method [46] |
Winner-takes-all method [46] |
Weighting method [46] |
|
Coverage optimization |
Ensemble selection |
Classifier level | Homogenous | Clustering-based selection method [17, 45, 88, 91] |
Threshold-based selection method [86] |
Heterogeneous | — |
Ensemble generation | Feature level | Feature selection/reduction | Random subspace method [46] |
The input decimation method [90] |
Genetic algorithms [92] |
Markov blanket BN [28] |
Principal component analysis [93] |
Information theory [16] |
Data level | Resampling | Bagging [61] |
Wagging [94] |
Random forest [95] |
Boosting [96] |
Stacking [79] |
Output code method | One per class (OPC) [97] |
Pairwise coupling [98] |
Correcting classifiers [99] |
Pairwise coupling correcting classifiers [99] |
Error-correcting output coding [100] |
Data-driven ECOC [101] |
|