Skip to main content

Table 1 Model type of the 152 models developed in the 62 included publications

From: Methodological conduct of prognostic prediction models developed using machine learning in oncology: a systematic review

Model characteristics All models (n = 152)
n (%)
Regression-based models 42 (28)
 Logistic regression 26
 Cox regression 7
 Linear regression 3
 LASSO (Logistic regression) 1
 LASSO (Cox regression) 1
 LASSO (model not specified) 3
 Best subset regression with leave-out cross-validation 1
Non-regression-based models 71 (47)
 Neural network (including deep learning) 18
 Classification tree (e.g., CART, decision tree) 28
 Support vector machine 12
 Naive Bayes 6
 K nearest neighbours 3
 Othera 4
Ensemble models 39 (26)
 Random forest (including random survival forest) 23
 Gradient boosting machine 8
 RUSBoost - boosted random forests 1
 Bagging with J48 selected by Auto-WEKA 1
 CoxBoost - boosted Cox regression 1
 XGBoost: exTreme Gradient Boosting 1
 Gradient boosting machine and Nystroem, combined using elastic net 1
 Adaboost 1
 Bagging, method not specified 1
 Partitioning Around Medoid algorithm and complete linkage method 1
Median number of models developed per study [IQR], range 2 [1–4], 1–6
  1. CART Classification And Regression Tree, LASSO Least Absolute Shrinkage and Selection Operator
  2. aOther includes voted perceptron; fuzzy logic, soft set theory and soft set computing; hierarchical clustering model based on the unsupervised learning for survival data using the distance matrix of survival curves; Bayes point machine