Table 2. Comparisons among the accuracy scores obtained by all the BN-based feature selection methods for the non-dominant leg.
In grey is highlighted the best performing BN.
| Feature selection algorithm | Search technique | AUC | N° of features selected | Description in ascending (from more to less important/relevant) order |
|---|---|---|---|---|
| - | - | 0.821 | 31 | S1 Table |
| Correlation-based feature subset evaluator | Best First | 0.817 | 8 | Dominant-leg, ISOM-Hip-Abd, CS-WF, CS-ML, ROM-HE, ROM-KF, ROM-AKDFKE and ROM-AKDFKF |
| Chi squared attribute evaluator | Ranker | 0.879 | 10 | ROM-AKDFKE, ROM-AKDFKF, ROM-KF, ROM-HE, CS-ML, CS-CD, CS-WF, ROM-HFKF, ISOK-ECC-KF180 and CS-NF |
| Classifier attribute evaluator (Naïve Bayes) | Ranker | 0.809 | 10 | ROM-AKDFKF, ROM-KF, ROM-HE, ISOK-ECC-KF180, ROM-AKDFKE, ROM-HFKF, CS-WF, ISOK-ECC-KE30, ISOK-ECC-KE60 and CS-CD |
| Classifier subset evaluator (Naïve Bayes) | Best First | 0.758 | 10 | ISOK-ECC-KF180, ISOK-ECC-KE60, ISOM-Hip-Add, CS-NF, CS-WF, CS-CD, ROM-HE, ROM-KF, ROM-AKDFKE and ROM-AKDFKF |
| Consistency subset evaluator | Best First | 0.828 | 5 | ROM-HABD, ROM-HIR, ROM-HER, ROM-KF and ROM-AKDFKF |
| Correlation attribute evaluator | Ranker | 0.853 | 9 | ROM-AKDFKE, ROM-AKDFKF, ROM-KF, CS-ML, ROM-HFKF, CS-WF, CS-NF, ISOM-Hip-Add and Dominant-leg |
| CV Attribute evaluator | Ranker | 0.700 | 9 | ROM-AKDFKE, Dominant-leg, ISOK-ECC-KF180, ISOK-ECC-KF60, ISOK-ECC-KE30, ISOK-CON-KE240, ISOK-ECC-KE60, ISOK-ECC-KF30 and ROM-AKDFKF |
| Gain ratio attribute evaluator | Ranker | 0.853 | 10 | ROM-AKDFKE, ROM-AKDFKF, ROM-KF, CS-ML, ROM-HFKF, CS-WF, CS-NF, Dominant-leg, ISOM-Hip-Add and ROM-HE |
| Info gain attribute evaluator | Ranker | 0.853 | 9 | ROM-AKDFKE, ROM-AKDFKF, ROM-KF, CS-ML, ROM-HE, CS-CD, ROM-HFKF, CS-WF, ISOK-ECC-KF180 and CS-NF |
| One R attribute evaluator | Ranker | 0.731 | 9 | ROM-AKDFKF, ROM-KF, ROM-HE, ISOK-ECC-KF180, ROM-AKDFKE, ISOK-ECC-KE60, ISOK-ECC-KF60, ISOK-CON-KF240 and ISOK-CON-KF180 |
| Wrapper subset evaluator (Naïve Bayes) | Best First | 0.809 | 22 | ISOK-CON-KF60, Body-mass, ISOK-CON-KE180, ISOK-CON-KE240, ISOK-ECC-KF30, ISOK-ECC-KF60, ISOK-ECC-KF180, ISOK-ECC-KE30, ISOK-ECC-KE60, ISOM-Hip-Abd, ISOM-Hip-Add, CS-NF, CS-ML, CS-AP, CS-CD, ROM-HFKF, ROM-HFKE, ROM-HABD, ROM-HE, ROM-KF, ROM-AKDFKE and ROM-AKDFKF |
BN: Bayesian Network Classifiers; AUC: area under the receiver operating characteristic curve; ISOK: isokinetic; KE: knee extensors; CON: concentric; ECC: eccentric; ISOM: isometric; PT: peak torque; Abd: abduction; ROM: range of motion; HFKF: hip flexion with the knee flexed; HE: Hip extension; HIR: hip internal rotation; HER: hip external rotation; KF: knee flexors; AKDFKE: ankle dorsi-flexion with the knee extended; AKDFKF: ankle dorsi-flexion with the knee flexed; CS: core stability; NF: unstable sitting without feedback; WF: unstable sitting with feedback; ML: unstable sitting while performing medial-lateral displacements with feedback; AP: unstable sitting while performing anterior-posterior displacements with feedback; CD: unstable sitting while performing circular displacements with feedback.