Тип публикации: статья из журнала
Год издания: 2021
Идентификатор DOI: 10.1007/978-3-030-70594-7_9
Ключевые слова: cardiovascular death, evolutionary search, fuzzy logic, population study, predictive modeling
Аннотация: Predictive models are commonly used in epidemiological studies to estimate risks of illnesses. For knowledge-based models, the logic behind is clear, whereas for automatically generated data-driven models, it is not always transparent how they work. In this study, we applied an evolutionary approach to design a Fuzzy Logic Rule-basПоказать полностьюed model that is easily interpretable compared to many other data-driven models. We utilized a high-dimensional epidemiological data collected within the Kuopio Ischemic Heart Disease Risk Factor (KIHD) Study in 1984–1989 to train the model and predict cardiovascular death for middle-aged men by 2016. In multiple runs of 5-fold cross-validation, we evaluated the model performance and showed that it could achieve higher true positive rate (TPR) than Random Forest and provide more stable results than Decision Tree. Also, the presented approach proved its effectiveness for high-dimensional samples: on the set of 653 predictors, we obtained 68% accuracy on average, whereas on the reduced set of 100 predictors, we could improve this result only up to 70%. Furthermore, this study introduces the most important predictor variables used in the generated Fuzzy Logic Rule-based model.
Журнал: Studies in Computational Intelligence
Выпуск журнала: Т. 922
Номера страниц: 223-236
ISSN журнала: 1860949X