Prediction and classification the risk of stroke patients using data mining techniques

Abstract

Background Stroke is one of the most common causes of death and neurological disabilities in all societies. The use of data mining techniques to create predictive models is very helpful in identifying people at risk to reduce the complications of the disease. The purpose of this study was to investigate the performance of data mining algorithms and predict the risk of stroke in suspected stroke patients using decision tree based on the risk factors that affect it. Methods The study analyzed medical records of 1184 stroke-suspected patients presenting at an Emergency Department using data mining algorithms. Attributes such as age, primary diagnoses, gender, blood pressure, smoking, diabetes, and other relevant factors were considered. The dataset was preprocessed to handle missing and incompatible data. Algorithms like Naïve Bayes, Neural Network, kNN, SVM, and Classification Tree were applied, with a training-test data split of 70 − 30 using K-fold Cross Validation. Results Among the data mining algorithms used, kNN demonstrated the highest accuracy (97.30%), sensitivity (98.75%), specificity (98.72%), and F1 criteria (98.66%) in predicting stroke severity. Physical inactivity, high cholesterol, cardiovascular disease, history of transient ischemic attack, and high blood pressure emerged as the most influential risk factors for stroke severity. Decision Tree analysis provided valuable insights into the relationship between risk factors and stroke severity. Conclusion Data mining techniques proved effective in identifying risk factors and predicting stroke severity, showing promise in enhancing stroke management strategies. The study highlighted the importance of physical inactivity and other key risk factors in stroke prediction. Consistency in risk factor importance across studies suggests common underlying factors, while acknowledging variations based on geographic and lifestyle factors.