TableĀ 3

Algorithm performance identifying comments relating to positive and negative experience: 50:50 split between training and testing samples approach and 10-fold cross validation

AlgorithmApproach 1: 50% comments used for training and 50% for testingApproach 2: 10-fold cross-validation
PrecisionSensitivityF-scoreMean performance score across 10-foldsSD
Support vector machines (SVM)0.8350.7800.8000.8340.027
Random forests0.8250.7650.7800.8390.028
Decision trees0.7350.7100.7200.7700.050
Generalised linear models network (GLMNET)0.7500.7000.7100.5230.084
Bagging0.7250.7000.7100.8110.039
Maxentropy0.6700.6700.6700.0140.009
Logitboost0.7100.6550.6550.8760.037
  • GLMNET and Bagging have the same F-score, but precision was higher for GLMNET.