Prediction of epileptic seizure using regression model

Abstract

Epilepsy seizure detection by means of machine learning and deep newlinelearning models is the emerging research field. There are certain works that newlineemploys machine learning algorithm for seizure detection. Thus the first phase newlineof this work concentrates on using the hyper parameter tuning to achieve better newlineresults. For enabling the best classifier, nine different classifiers have been newlineexperimented and compared with one another in terms of accuracy and other newlineperformance parameters. The classifiers experimented with are K-Nearest newlineNeighbor, Decision Tree, Naïve Bayes, Random forest, Stochastic Gradient newlineDescent, Gradient Boosting, XGBoost, Extra Tree Classifier, and Logistic newlineregression.The best performing machine learning classifier is identified and fine newlinetune for increasing the performance. From the obtained results, it is noticed that newlinethe Extra Tree classifier performs better than the other models. The accuracy newlineobtained with it is 96.5%. Modification in several features and Hyperparameter newlinetuning is done to increase the classifier performance using a genetic algorithm newlineapproach. When experimented with the Extra tree classifier s max features newlineproperty, the best accuracy obtained is 98%. The performance of the model increases considerably when the genetic algorithm based AutoML approach is used for hyperparameter tuning. In addition to this a convolutional neural network model is also designed for seizure detection and the results are compared. The CNN model is found to perform better than the extra tree classifier and the result obtained is 99.6%, newlinewhich is higher than the other state of art models. newline newline