Medical Diagnosis Models Of Stroke Prediction Using Electronic Health Records

Abstract

Stroke is one of the most serious life-threatening medical conditions that occurs when the blood supply to part of the brain is cut-off. Stroke itself has so many sequelae. It can lead to lasting brain damage, long-term disability, or even more dangerous sequelaes including death. Despite the advances in medical technology, early detection and prediction of stroke remain difficult due to its complex nature and multitude of risk factors involved. This thesis presents a novel approach to the field of stroke prediction with the aim of applying machine learning techniques for electronic health records (EHR) in order to determine the early symptoms of stroke. The main focus of this research is about data preprocessing techniques which may be useful for the EHR data and improve the performance of machine learning models in this type of data. This research applies the domain knowledge of stroke in conjunction with various data preprocessing techniques to the electronic health records (EHR) data. The processing techniques include using domain knowledge in the medical field about stroke to extract the most important features from the dataset, encoding categorical variables, categorizing the data, handling missing values, among others. These techniques are crucial in the part of preparing the data for effective modeling and ensuring the reliability of the studies. In order to check the effectiveness of the preprocessing techniques, a variety of machine learning models are used to evaluate the performance. These models span different categories, including tree-based models like Decision Trees and Random Forests, distance-based models such as K-Nearest Neighbors (KNN), and probabilistic models like Naive Bayes. Each model’s performance is assessed using metrics including accuracy, precision, recall, and F1-Score. This comprehensive evaluation allows us to understand the impact of preprocessing techniques on the performance of different types of models. The findings provide valuable insights into the role of data preprocessing in stroke prediction and can guide future research in this area. It is believed that the approach can be generalized to other medical conditions, paving the way for the development of robust and reliable predictive models in healthcare. Future work will involve exploring more advanced preprocessing techniques and machine learning models to further improve stroke prediction.