Table 10.
Comparison of the proposed method with some prior studies using the Cleveland dataset (the best value is indicated in bold).
| Study | Technique | # of features | Validation method | Accuracy (%) | F 1 (%) | AUC (%) | Statistical test |
|---|---|---|---|---|---|---|---|
| [13] | Rotation forest-J48-CFS | 7 | 10CV | 84.48 | Not reported | 89.5 | No |
| [14] | PSO fuzzy expert systems | 76 | Hold-out | 93.27 | Not reported | Not reported | No |
| [15] | SMO-expert-based feature selection | 8 | 10CV | 84.49 | 86.2 | Not reported | No |
| [19] | CFS-PSO-clustering-MLP | 5 | 10CV | 90.28 | Not reported | Not reported | No |
| [22] | Logistic regression-LASSO | 6 | 10CV | 89 | Not reported | Not reported | No |
| [24] | Boosted-C5.0 and neural network | 12 | 10CV | 77.8 & 81.9 | Not reported | Not reported | Paired t test |
| [9] | Voting-naive Bayes-logistic regression | 9 | 10CV | 87.41 | Not reported | Not reported | No |
| This paper | Two-tier ensemble PSO-based feature selection | 7 | 10CV | 85.71 | 86.49 | 85.86 | Two-step statistical test |