Skip to main content

Table 5 Results of stratified 10 × 5-fold cross validation with the CV set (patients recruited in Lisbon, Table 3), under the Time Windows and the First Last approaches

From: Predicting progression of mild cognitive impairment to dementia using neuropsychological data: a supervised learning approach using time windows

  AUC Sensitivity Specificity
  FL 2Y 3Y 4Y 5Y FL 2Y 3Y 4Y 5Y FL 2Y 3Y 4Y 5Y
DT 0.65 ± 0.02 0.71 ± 0.04 0.75 ± 0.01 0.78 ± 0.02 0.79 ± 0.02 0.59 ± 0.03 0.65 ± 0.05 0.75 ± 0.04 0.71 ± 0.04 0.77 ± 0.02 0.68 ± 0.02 0.73 ± 0.04 0.69 ± 0.03 0.77 ± 0.03 0.75 ± 0.02
kNN 0.67 ± 0.01 0.77 ± 0.01 0.82 ± 0.01 0.83 ± 0.01 0.84 ± 0.01 0.60 ± 0.02 0.70 ± 0.03 0.87 ± 0.01 0.69 ± 0.03 0.83 ± 0.01 0.65 ± 0.01 0.71 ± 0.01 0.61 ± 0.01 0.81 ± 0.02 0.72 ± 0.03
SVM Poly 0.63 ± 0.01 0.70 ± 0.01 0.76 ± 0.01 0.79 ± 0.01 0.80 ± 0.01 0.43 ± 0.02 0.55 ± 0.02 0.71 ± 0.01 0.81 ± 0.02 0.86 ± 0.01 0.83 ± 0.01 0.84 ± 0.01 0.81 ± 0.01 0.77 ± 0.01 0.75 ± 0.02
SVM RBF 0.63 ± 0.01 0.64 ± 0.01 0.76 ± 0.01 0.79 ± 0.01 0.80 ± 0.02 0.40 ± 0.02 0.35 ± 0.02 0.72 ± 0.02 0.80 ± 0.02 0.89 ± 0.01 0.86 ± 0.01 0.93 ± 0.01 0.81 ± 0.01 0.78 ± 0.02 0.71 ± 0.03
NB 0.74 ± 0.00 0.82 ± 0.01 0.86 ± 0.00 0.87 ± 0.01 0.88 ± 0.00 0.64 ± 0.02 0.66 ± 0.01 0.75 ± 0.02 0.82 ± 0.01 0.88 ± 0.01 0.71 ± 0.02 0.82 ± 0.01 0.79 ± 0.01 0.78 ± 0.01 0.71 ± 0.01
LR 0.72 ± 0.01 0.79 ± 0.01 0.84 ± 0.01 0.84 ± 0.01 0.85 ± 0.01 0.47 ± 0.01 0.77 ± 0.02 0.85 ± 0.03 0.74 ± 0.01 0.78 ± 0.01 0.80 ± 0.01 0.66 ± 0.01 0.68 ± 0.02 0.81 ± 0.02 0.78 ± 0.02
RF 0.72 ± 0.01 0.79 ± 0.01 0.85 ± 0.01 0.86 ± 0.01 0.87 ± 0.01 0.59 ± 0.03 0.53 ± 0.04 0.75 ± 0.01 0.75 ± 0.01 0.87 ± 0.01 0.71 ± 0.02 0.86 ± 0.01 0.77 ± 0.02 0.81 ± 0.01 0.70 ± 0.02
  1. Note: DT: Decision Tree classifier, kNN: k-Nearest Neighbor classifier, SVM Poly: polynomial-kernel Support Vector Machines, SVM RB: Gaussian-kernel Support Vector Machines, NB: Naïve Bayes classifier, LR: Logistic Regression and RF: Random Forest
  2. The results were highlighted in bold whenever Time Windows approach outperformed the FL approach. cMCI represents the positive class