Deteksi Sinyal P300 Menggunakan Metode Batch Normalization Neural Network = P300 Signal Detection Using Batch Normalization Neural Network

Purnomo, Adenuar (2018) Deteksi Sinyal P300 Menggunakan Metode Batch Normalization Neural Network = P300 Signal Detection Using Batch Normalization Neural Network. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

[img] Text
05111440000079-Undergraduate_Theses.pdf - Published Version
Restricted to Repository staff only

Download (9MB) | Request a copy

Abstract

Brain-computer interface (BCI) adalah mekanisme komunikasi antara sinyal electroencephalograph (EEG) dan komputer dimana sehingga sistem dapat menangkap niat otak tanpa melibatkan neuron motorik dan muskular. Mendeteksi sinyal P300 dari EEG adalah kunci untuk menafsirkan maksud dari pengguna. Convolutional Neural Network (CNN) dapat mendeteksi sinyal P300 dengan baik, namun CNN cenderung overfitting. Untuk mengatasi masalah ini diimplementasikan sebuah metode yang bernama Batch Normalisation Neural Network (BNNN) untuk mendeteksi sinyal P300. Sinyal inputan pertama-tama dipotong terlebih dahulu sepanjang 0-667ms setelah stimulus, setelah itu difilter menggunakan Butterworth Filter. Jumlah data masing-masing kelas pada data inputan tidak seimbang, sehingga dilakukan duplikasi untuk menyeimbangkannya. Kemudian dilakukan normalisasi menggunakan min-max normalization, dan dilakukan signal averaging. Data kemudian diperhalus menggunakan Haar Wavelet Transformation dan inverse Haar Wavelet Transformation. Barulah data dapat diklasifikasi menggunakan Batch Normalisation Neural Network. Performa metode klasifikasi yang diusulkan diuji dengan membandingkan kelas sinyal hasil prediksi dengan kelas sinyal ground truth menggunakan perhitungan akurasi, specificity, dan sensitivity. Dilakukan beberapa uji coba untuk mencari parameter terbaik dari jumlah sinyal yang di averaging, level smoothing, jumlah epoch, nilai dropout probability, jumlah kernel fully connected layer, dan jumlah kernel dari convolutional layer. Dari beberapa skenario uji coba didapatkan hasil akurasi terbaik sebesar 70.80% untuk data BCI competition 2 dataset 2B, 84.03% untuk data BCI competition 3 dataset 2 subjek A dan 86.85% untuk data BCI competition 3 dataset 2 subjek B. Dari hasil ujicoba tersebut, BNNN dapat memberikan hasil yang baik untuk beberapa dataset. ============ Brain -computer interface (BCI) is a communication mechanism between electroencephalograph (EEG) signals and a computer, such that the system can capture the brain intention without involving motoric and muscular neurons. Detecting the P300 signal from electroencephalograph (EEG) is the key to interpreting the intent of the user. Convolutional Neural Network (CNN) can detect P300 signal well, but CNN tends to overfitting. To solve this problem, a method called Batch Normalization Neural Network (BNNN) is implemented to detect P300 signals. The input signal is firstly cut 0-667ms after the stimulus, after that input signal is filtered using the Butterworth Filter. The amount of data each class on the input data is not balanced, so duplication is done to balance it. Then normalization is performed using min-max normalization, after that averaging signal is performed. The data is then refined using Haar Wavelet Transformation and inverse Haar Wavelet Transformation. Only then can the data be classified using the Batch Normalisation Neural Network. The performance of the proposed classification method is tested by comparing the predicted signal class with the ground truth signal class using accuracy, specificity, and sensitivity calculations. There are several tests to find the best parameters of the number of averaging signal, smoothing level, epoch number, dropout probability value, number of kernel fully connected layer, and number of kernels from the convolutional layer. From several test scenarios, the best accuracy result is 70 .80 % for BCI competition 2 dataset 2B data, 84.03% for BCI competition 3 dataset 2 subject A and 86.85% for BCI competition 3 dataset 2 subject B. From the result of the test, BNNN can give good results for some datasets

Item Type: Thesis (Undergraduate)
Additional Information: RSIf 005.133 Pur d-1
Uncontrolled Keywords: EEG; P300 event-related potential; CNN; BNNN
Subjects: Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
R Medicine > RC Internal medicine > RC0321 Neuroscience. Biological psychiatry. Neuropsychiatry
T Technology > T Technology (General) > T57.5 Data Processing
Divisions: Faculty of Information and Communication Technology > Informatics > (S1) Undergraduate Theses
Depositing User: Purnomo Adenuar
Date Deposited: 08 Jan 2019 07:24
Last Modified: 08 Jan 2019 07:24
URI: http://repository.its.ac.id/id/eprint/53512

Actions (login required)

View Item View Item