Table 3.
Classification and Prediction of UHR’s individuals’ outcomes Using Machine Learning Tools.
| (a) Classifying between Remitter and Maintained | ||||||
|---|---|---|---|---|---|---|
| Algorithm | Accuracy % | F1 | MCC | AUROC | Sensitivity | Specificity |
| LSTM | 0.593 Sd (0.07) | 0.231 (0.17) | 0.019 (0.16) | 0.493 (0.08) | 0.787 (0.16) | 0.228 (0.19) |
| CNN | 0.524 (0.12) | 0.294 (0.18) | 0.012 (0.12) | 0.498 (0.06) | 0.628 (0.34) | 0.384 (0.35) |
| SVM | 0.625 (0.07) | 0.253 (0.14) | 0.063 (0.17) | 0.524 (0.07) | 0.839 (0.09) | 0.219 (0.31) |
| Random Forest | 0.596 (0.09) | 0.220 (0.16) | 0.047 (0.21) | 0.518 (0.07) | 0.868 (0.10) | 0.166 (0.31) |
| (b) Predicting between Remitter and Maintained (Prognosis) Baseline to 18 months data and labels at 24 months | ||||||
|---|---|---|---|---|---|---|
| Algorithm | Accuracy | F1 | MCC | AUROC | Sensitivity | Specificity |
| LSTM | 0.569 (0.07) | 0.175 (0.14) | −0.050 (0.12) | 0.474 (0.08) | 0.783 (0.17) | 0.176 (0.18) |
| CNN | 0.556 (0.09) | 0.291 (0.15) | 0.005 (0.12) | 0.492 (0.07) | 0.664 (0.24) | 0.335 (0.25) |
| SVM | 0.641 (0.07) | 0.280 (0.17) | 0.079 (0.20) | 0.533 (0.08) | 0.825 (0.10) | 0.241 (0.16) |
| Random Forest | 0.633 (0.08) | 0.288 (0.13) | 0.114 (0.19) | 0.541 (0.07) | 0.842 (0.12) | 0.241 (0.13) |
(a) Classification results of two classes (Remitter and Maintained) using machine learning tools including Random Forest, Support vector machine (SVM), K-nearest neighbour (KNN), and Long Short term memory (LSTM) and Convolutional neural network (CNN).
(b) The prediction task was also conducted using Random Forest, SVM, KNN, LSTM, and CNN. The F score, Matthew’s correlation coefficient (MCC), Area Under the Curve (AUROC) for measuring the performance of the machine learning tools are also reported.