Table 3.
Classification studies employing biomechanical data and/or distinct variables.
| Author | Year | Data | Feature engineering | Learning Algorithm | Validation | Results |
|---|---|---|---|---|---|---|
| Aksehirli, Ö [92] |
2013 | Demographic characteristics and some gene polymorphisms | – | SVM, PNN |
152 OA knees for training and 102 healthy for testing | 76,77% acc & 90,55% acc |
| Beynon, M. J. [78] |
2006 | Biomechanical Data | Simulated annealing (SA) and genetic algorithms (GAs) |
Dempster–Shafer theory of evidence (DST) & Linear discriminant analysis (LDA) | LOOCV | 96.7% & 93.3% acc |
| de Dieu Uwisengeyimana, J [89]. | 2017 | Biomechanical Data | Time-domain statistical features | Multilayer perceptron, Quadratic support vector machine, complex tree & deep learning network with k-NN |
22 subjects (11 healthy and 11 OA) | 99.5%, 99.4% 98.3% & 91.3% acc |
| Deluzio, K.J. [84] |
2007 | Biomechanical Data | PCA | Discriminant analysis | CV | Misclassification rate 8% |
| Jones, L. [85] |
2008 | Biomechanical Data | PCA | The Dempster–Shafer (DS)-based classifier & ANN |
LOOCV | 97.62% & 77.82% acc |
| Kotti, M. [83] |
2014 | Biomechanical data | PPCA | Bayes classifier | 47F-CV | 82.62% acc |
| Kotti, M. [90] |
2017 | Biomechanical data | – | Random forest | 50% training/50% testing, 5F-CV | 72.61% acc |
| Lim J. [86] |
2019 | Demographic and personal characteristics, lifestyle- and health status-related variables | PCA | DNN | 66% training/34% testing | AUC of 76.8% |
| Long, M. J. [94] |
2017 | Outcome scores and biomechanical gait parameters | – | KNN | 70% training/30% test. 30% of training was left out for validation | AUC of 1.00 |
| McBride, J. [95] |
2011 | Biomechanical data | – | Neural networks | 50% training/50% testing | 75.3% acc |
| Mezghani, N. [79] |
2008 | Biomechanical data | Discrete wavelet transform (DWT) | Nearest neighbor classification (NNC) | LOOCV | 38 of 42 cases acc |
| Mezghani, N. [80] |
2008 | Biomechanical data | Discrete wavelet transform (DWT) & Polynomial expansion |
Nearest neighbor classifier (NNC) | LOOCV | 91% acc 67% acc |
| Mezghani N. [96] |
2017 | Biomechanical Data | – | Regression tree | 10F-CV for model selection. 10% for model evaluation | ROC AUC of 0.85 |
| Moustakidis, S. [81] |
2010 | Biomechanical data | Wavelet Packet, FS via SVMFuzCoC | KNN1 SVM (AAA) SVM (1AA) FCT C4.5 FDT-SVM |
10F-CV | 86.09% acc 89.71% acc 90.18% acc 88.35% acc 91.12% acc 93.44% acc |
| Moustakidis, S. [88] |
2019 | Clinical Data | Feature subsets exploration | DNN Adaboost Fuzzy KNN Fuzzy NPC CFKNN |
10F-CV | 86.95% acc (for age 70+) 78.60% acc 77.39% acc 72.40% acc 73.60% acc |
| Phinyomark, A. [87] |
2016 | Biomechanical Data | PCA | SVM | 10F-CV | 98–100% acc |
| Şen Köktaş, N. [93] |
2006 | Biomechanical data | – | MLPs | CV | 1.5 of the subjects has been misclassified |
| Şen Köktaş, N. [82] |
2010 | Biomechanical data (Also included age, body mass index and pain level) |
Mahalanobis Distance algorithm | Decision tree - MLP multi-classifier | 10F-CV | 80% acc |
| Yoo, T. K. [91] |
2016 | Predictors of the scoring system in the Fifth Korea National Health and Nutrition Examination Surveys (KNHANES V-1) data | Logistic regression | ANN | 66.7% training/33.3% validation, KNHANES V-1 (internal validation group) and OAI (external validation group) |
ROC AUC of 0.66–0.88 |