The Application Of Random Forest And KNN Model In EEG Data Based On Outlier Processing

Recently, with the rapid development of deep learning and artificial intelligence,researchers begin to use these new technologies to study the issue about EEG. Using the EEG, doctors are better able to diagnose brain disorders and researchers can understand the relationship between brain waves and behavioral activity to develop more intelligent devices. In this paper, we use EEG data collected by EEG measurement instrument as input, and regard the corresponding status of eye state as output. To improve the reliability and accuracy of classification on this data, Random Forest model and kNN model are used to model the classification problem after deleting the outliers according to the property and the changing characteristics of the data. This article firstly applies data preprocessing in raw data, mainly includes missing values processing, outlier processing and consistency analysis. For the data used in this paper, we use statistical analysis and visualization to deal with the outliers. Secondly, using Random Forest and kNN to build the specific models after deleting outliers. For Random Forest model, we also discuss the OOB error rate of the method and the importance of variables in the model; for kNN model, cross validation method is used to determine the k value on the train set, then we use it to evaluate the prediction effect on test data set. Finally, discussing the result. In order to show the validity of Random Forest model and kNN model, we use Decision Tree,Bagging and SVM as the comparing approaches. At the same time, we also discussed the effect of the imbalance of sample in the data set on the model. Results show that the proposed Random Forest model and kNN model based on outliers processing have better prediction accuracy. The prediction accuracy of Random Forest model reaches 92.9392%,and the prediction accuracy of kNN model reaches 97.0946%. So we can get the conclusion that Random Forest model and kNN model are effective in this EEG data set, especially kNN model, compared with other methods in this paper, has the best effect on prediction.