Skip to main content

Table 4 Evaluation results of each model using screening data in 2016

From: Using machine learning models to improve stroke risk level classification methods of China national stroke screening

Learning methodPrecision (95% CI)Recall (95% CI)F1-score (95%CI)AUC (95% CI)
Logistic regression90.56% [90.52,90.60%]96.35% [96.31,96.39%]93.37% [93.33,93.41%]97.96% [99.09,99.19%]
Naïve Bayesian66.96% [66.93,66.99%]94.99% [94.95,95.03%]78.55% [78.51,78.58%]96.64% [96.62,96.66%]
Bayesian network67.50% [67.47,67.53%]93.85% [93.80,93.90%]78.52% [78.49,78.56%]96.86% [96.82,96.90%]
Decision tree(C4.5)91.95% [91.90,92.00%]98.12% [98.09,98.15%]94.93% [94.89,94.98%]99.36% [99.33,99.39%]
Neural network91.82% [91.78,91.86%]98.52% [98.49,98.55%]95.05% [95.02,95.09%]99.23% [99.20,99.26%]
Random forest96.89% [96.86,96.92%]95.76% [95.74,95.78%]96.32% [96.30,96.35%]99.41% [99.39,99.43%]
Bagging with C4.5 decision tree92.21% [92.19,92.23%]98.86% [98.83,98.89%]95.42% [95.39,95.44%]99.39% [99.92,99.94%]
Voting92.12% [92.07,92.17%]98.98% [98.96,99.00%]95.43% [95.39,95.46%]99.39% [99.36,99.42%]
Boosting with C4.5 decision tree94.89% [94.85,94.93%]99.12% [99.09,99.15%]96.96% [96.92,96.99%]99.41% [99.38,99.44%]
  1. *The bold in tables is the maximum value of that evaluation standard