چكيده به لاتين
Stroke is one of the most important and common neurological diseases and it affects millions of people around the world every year. Due to the increase in death rates due to stroke and also the high costs that this complication imposes on health organizations, all countries have made prevention one of the highest priorities. Considering the development of using different data mining methods in presenting stroke prediction models, it seems necessary to know and compare these methods.
The present study was conducted in 2021 with the aim of providing a model based on statistical / machine learning approaches for predicting stroke. The novelty of the present study is the use of 9 different algorithms which, compared to previous researches, led to higher accuracy of stroke prediction.
For this purpose, the data of a data set, which included data related to 47557 people, was used. RapidMiner software was used to analyze the data. In order to present the stroke prediction model, the algorithms "Decision Tree, Naive Bayes, Role Induction, Neural Network, SVM, Logistic Regression, KNN, Random Forest and Hybrid Model" were used. Based on the research findings, the accuracy of different models was as follows: decision tree model (total accuracy= 90.63), Naive Bayes model (total accuracy= 88.86), Role induction model (total accuracy = 91/02), neural network model (total accuracy=91/02), SVM model (total accuracy=37.89), Logistic regression model (total accuracy = 83.40), KNN model (Total accuracy = 84/62), Random Forest model (total accuracy = 97/83) and Combined model of KNN, Naïve Bayes and rule induction (total accuracy=92.87). Based on the research findings, stroke prediction models based on "neural network, decision tree, hybrid model and Naïve Bayes" had the best performance.
