Skip to main content

Table 3 Key performance statistics of the trained models

From: Predicting hospitalization following psychiatric crisis care using machine learning

ML algorithm

AUC

Sensitivity

Specificity

Accuracy

Gradient boosting

0.774

0.455

0.894

0.744

Oblique random forest

0.762

0.509

0.847

0.732

DeepBoost

0.760

0.461

0.871

0.731

Random forest

0.757

0.478

0.864

0.732

GLM (logistic regression)

0.756

0.444

0.876

0.729

Support vector machines

0.751

0.370

0.917

0.731

Naive Bayes

0.751

0.455

0.861

0.723

Neural network

0.749

0.528

0.828

0.726

Keras/TensorFlow

0.741

0.465

0.850

0.719

K-nearest neighbors

0.702

0.356

0.879

0.701

  1. The base rate of (non-)hospitalization = 0.659. The accuracy of each model was tested against this base rate, all p < 0.00001, based on 2-sided z-tests; hence each model led to a significant improvement in classification accuracy compared to an intercept only model