Five layered Ensembled Deep Fully Connected Neural Network based Brain Stroke Prediction

Brain stroke infarction is triggered due to the blocking of blood passage in the brain. Researchers are experimenting the various genetic factors that influences the stroke. However, the exact identification of brain stroke remains a major challenge for the doctors. With this motivation, this paper aims to predict the existence of brain stroke by proposing Five Layered Ensemble Deep Fully Connected Neural Network (FLED) that uses the brain stroke database from the KAGGLE. The brain stroke sample has 10 characteristics and 4981 patient records that were encoded and treated with incomplete data. The exploratory data analysis have done to visualize the feature relations with the target. The dataset had applied to KNN, SVM, Decision tree, Random forest, XGB, ada boost, Gradient boost to analyze the performance of brain stroke prediction. Using a single input and output layer, the suggested Five Layered Ensemble Deep Fully Connected Neural Network was developed along with three hidden layers. The ensemble voting classifier had designed with five layered DFCNN, Ada Boost, Random Forest, Gradient Boosting and SVM to create an Ensemble Voting classifier and the performance was analyzed with all the classifiers. The proposed model Five Layered Ensemble Deep Fully Connected Neural Network have applied to the dataset to estimate the presence of brain stroke. Python was utilized for execution, and a server with a Geforce V100 Nvidia GPU was employed with a total sample size of 64 and 30 training iterations for analyzing the performance metrics. Experimental results shows that the proposed FLED model shows mean absolute error of 0.03745 and accuracy of 96.84% when compared with other classifier models.