|
| References | Prediction algorithm | Technique | Performance Metric | Advantages | Disadvantages |
|
| [16] | Workload prediction | MLP, MLPWD, SVM | MAE, RMSE, PRED, R2PA | Uses best prediction algorithm automatically | Uses the workload as predictor |
| [18] | SLA violation prediction | Regression and MPE | Number of violations | Reduced impact of SLA violations | Low violation prevention rate (78% maximum) |
| [12] | Predict resource demand | SVM, NN, LR | MAPE, RMSE, PRED | Resources need prediction without workload as predictor | Complex system with 11 input parameters |
| [19] | Resource usage prediction | Bayesian information | MAE, RMSE, MAPE | Prediction based on resource usage history | Only applied to CPU load |
| [21] | Workload clustering, resource usage prediction | K-means, Bayesian learning | Delay, cost, SLA-violation rate, energy consumption | Considers SLA cost on the decision-making process | The prediction only takes one performance metric (response time) |
| [22] | Resource usage prediction | Fuzzy C-means, gray wolf | Energy consumption, execution time and cost, SLA violation and failure rates | Does not use the workload as the predictor | Two-stage prediction algorithm (clustering and GWO) |
| [20] | Resource need prediction | MVA | Unknown ([20] does not give enough information about performance metrics and evaluation) | Does not use workload as predictor | The prediction only takes one performance metric (response time) |
|
