Accurate Prediction Of Stroke Through Concatenated Gated Recurrent Unit And Adaboost Convolutional Neural Network

Abstract

Stroke, the greatest threat causes an enormous number of death. The medical field employs a range of data mining tools to help with early illness detection and diagnosis. Machine Learning (ML) methods have gained popularity in the prediction, diagnosis, evaluation of this illness; however, because the data are gathered from multiple institutions, there are problems with data quality. The research objective is to enhance the accuracy of stroke data by applying a better pre-processing technique and enhance the prediction of stroke using hybrid adaboost convolutional neural network. An improved method for determining possible risk factors and forecasting the probability of stroke is done using an open access clinical dataset. The dataset has less precise categorization and an excess fitting issue. To minimize the expense of the traditional AdaBoost when working with large sets of training data, an AdaBoost-Convolutional Neural Network (AB-CNN) was developed. The AB-CNN was implemented with minimum number of learning epochs for categorization of different classes of stroke. The proposed work analyses both modifiable and non-modifiable risk factors. Modifiable factors include lifestyle habits like smoking and medical conditions such as high blood pressure, while non-modifiable factors include age, gender, and family history. A combined deep learning model using Gated Recurrent Units (GRUs) and Convolutional Neural Networks (CNNs) is employed to capture patterns from both time-based and spatial data. This approach supports both MRI and clinical data. Thereby the proposed technique reduces the overall processing time and identifies individuals at high risk, enabling healthcare providers to offer timely preventive care.