Garuda - Garba Rujukan Digital

Post-Stroke identification of EEG signals using recurrent neural networks and long short-term memory Wanodya Sansiagi; Esmeralda Contessa Djamal; Daswara Djajasasmita; Arlisa Wulandari
International Journal of Advances in Intelligent Informatics Vol 7, No 2 (2021): July 2021
Publisher : Universitas Ahmad Dahlan

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.26555/ijain.v7i2.512

Stroke often causes disability, so patients need rehabilitation for recovery. Therefore, it is necessary to measure its effectiveness. An Electroencephalogram (EEG) can capture the improvement of activity in the brain in stroke rehabilitation. Therefore, the focus is on the identification of several post-rehabilitation conditions. This paper proposed identifying post-stroke EEG signals using Recurrent Neural Networks (RNN) to process sequential data. Memory control in the use of RNN adopted Long Short-Term Memory. Identification was provided out on two classes based on patient condition, particularly "No Stroke" and "Stroke". EEG signals are filtered using Wavelet to get the waves that characterize a stroke. The four waves and the average amplitude are features of the identification model. The experiment also varied the weight correction, i.e., Adaptive Moment Optimization (Adam) and Stochastic Gradient Descent (SGD). This research showed the highest accuracy using Wavelet without amplitude features of 94.80% for new data with Adam optimization model. Meanwhile, the feature configuration tested effect shows that the use of the amplitude feature slightly reduces the accuracy to 91.38%. The results also show that the effect of the optimization model, namely Adam has a higher accuracy of 94.8% compared to SGD, only 74.14%. The number of hidden layers showed that three hidden layers could slightly increase the accuracy from 93.10% to 94.8%. Therefore, wavelets as extraction are more significant than other configurations, which slightly differ in performance. Adam's model achieved convergence in earlier times, but the speed of each iteration is slower than the SGD model. Experiments also showed that the optimization model, number of epochs, configuration, and duration of the EEG signal provide the best accuracy settings.

Identification of Speed and Unique Letter of Handwriting Using Wavelet and Neural Networks Esmeralda Contessa Djamal
Proceeding of the Electrical Engineering Computer Science and Informatics Vol 2: EECSI 2015
Publisher : IAES Indonesia Section

Handwriting stroke reflects the personality and emotional condition. Graphology is scientific method to evaluation personality through handwriting. There are many features in graphology to identify personality. Several previous researches used page margins, spacing, baseline, vertical zone, font size, and the type of unique letter t. Other research also identify the personality of signatures. This research uses feature writing speed and the type of letters a, d, i m, and t to identify personalities using structural analysis and artificial neural networks. To improve accuracy, image writing extracted using wavelet transform. The system is built with the approach of the structure and symbol has been implemented in software. The results show a unique type of letter recognition by 74%, and the speed feature by 60% recognition. Variations training data greatly affect recognition results.

Paraphrase Detection Using Manhattan's Recurrent Neural Networks and Long Short-Term Memory Achmad Aziz; Esmeralda Contessa Djamal; Ridwan Ilyas
Proceeding of the Electrical Engineering Computer Science and Informatics Vol 6: EECSI 2019
Publisher : IAES Indonesia Section

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/eecsi.v6.1973

Natural Language Processing (NLP) is a part of artificial intelligence that can extract sentence structures from natural language. Some discussions about NLP are widely used, such as Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) to summarize papers with many sentences in them. Siamese Similarity is a term that applies repetitive twin network architecture to machine learning for sentence similarity. This architecture is also called Manhattan LSTM, which can be applied to the case of detecting paraphrase sentences. The paraphrase sentence must be recognized by machine learning first. Word2vec is used to convert sentences to vectors so they can be recognized in machine learning. This research has developed paraphrase sentence detection using Siamese Similarity with word2vec embedding. The experimental results showed that the amount of training data is dominant to the new data compared to the number of times and the variation in training data. Obtained data accuracy, 800,000 pairs provide accuracy reaching 99% of training data and 82.4% of new data. These results are better than the accuracy of the new data, with half of the training data only yielding 64%. While the amount of training data did not effect on training data.

Dynamic Hand Gesture Recognition Using Temporal-Stream Convolutional Neural Networks Fladio Armandika; Esmeralda Contessa Djamal; Fikri Nugraha; Fatan Kasyidi
Proceeding of the Electrical Engineering Computer Science and Informatics Vol 7, No 1: EECSI 2020
Publisher : IAES Indonesia Section

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/eecsi.v7.2048

Movement recognition is a hot issue in machine learning. The gesture recognition is related to video processing, which gives problems in various aspects. Some of them are separating the image against the background firmly. This problem has consequences when there are incredibly different settings from the training data. The next challenge is the number of images processed at a time that forms motion. Previous studies have conducted experiments on the Deep Convolutional Neural Network architecture to detect actions on sequential model balancing each other on frames and motion between frames. The challenge of identifying objects in a temporal video image is the number of parameters needed to do a simple video classification so that the estimated motion of the object in each picture frame is needed. This paper proposed the classification of hand movement patterns with the Single Stream Temporal Convolutional Neural Networks approach. This model was robust against extreme non-training data, giving an accuracy of up to 81,7%. The model used a 50 layers ResNet architecture with recorded video training.

Semantic Classification of Scientific Sentence Pair Using Recurrent Neural Network Agung Besti; Ridwan Ilyas; Fatan Kasyidi; Esmeralda Contessa Djamal
Proceeding of the Electrical Engineering Computer Science and Informatics Vol 7, No 1: EECSI 2020
Publisher : IAES Indonesia Section

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/eecsi.v7.2051

One development of Natural Language Processing is the semantic classification of sentences and documents. The challenge is finding relationships between words and between documents through a computational model. The development of machine learning makes it possible to try out various possibilities that provide classification capabilities. This paper proposes the semantic classification of sentence pairs using Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM). Each couple of sentences is turned into vectors using Word2Vec. Experiments carried out using CBOW and Skip-Gram to get the best combination. The results are obtained that word embedding using CBOW produces better than Skip-Gram, although it is still around 5%. However, CBOW slows slightly at the beginning of iteration but is stable towards convergence. Classification of all six classes, namely Equivalent, Similar, Specific, No Alignment, Related, and Opposite. As a result of the unbalanced data set, the retraining was conducted by eliminating a few classes member from the data set, thus providing an accuracy of 73% for non-training data. The results showed that the Adam model gave a faster convergence at the start of training compared to the SGD model, and AdaDelta, which was built, gave 75% better accuracy with an F1-Score of 67%.

Classification of Post-Stroke EEG Signal Using Genetic Algorithm and Recurrent Neural Networks Ella Wahyu Guntari; Esmeralda Contessa Djamal; Fikri Nugraha; Sandi Lesmana Liem
Proceeding of the Electrical Engineering Computer Science and Informatics Vol 7, No 1: EECSI 2020
Publisher : IAES Indonesia Section

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/eecsi.v7.2052

Stroke is caused by a sudden burst of blood vessels in the brain, causing speech difficulties, memory loss, and also paralysis. The identification of electrical activity in the brain of post-stroke patients from EEG signals is an attempt to evaluate rehabilitation. EEG signal recording involves multiple channels with overlapping information. Therefore the importance of channel optimization is to reduce processing time and reduce the computational burden. Besides, that channel optimization can have an overfitting effect due to excessive utilization of EEG channels. This paper proposed the optimization of EEG channels for the identification of post-stroke patients using Genetic Algorithms and Recurrent Neural Networks. Data was taken from 75 subjects with a recording duration of 180 seconds in a seated state. The data was segmented and extracted using Wavelet to get the frequency of the Alpha, Theta, Mu, Delta, and Amplitude changes. The next step is the channel optimization process using Genetic Algorithms. The method applied to get a combination of channels that qualifies. Then, the EEG signal identification proceeds of the optimization of the channels used Recurrent Neural Network. The result showed that applying the Genetic Algorithm afforded 12 channels configuration with 90.00% of accuracy; meanwhile, used all channels gave a 72.22% result. Therefore, channel optimization is essential to reduce redundancy and increase recognition.

Hand Movement Identification Using Single-Stream Spatial Convolutional Neural Networks Aldi Sidik Permana; Esmeralda Contessa Djamal; Fikri Nugraha; Fatan Kasyidi
Proceeding of the Electrical Engineering Computer Science and Informatics Vol 7, No 1: EECSI 2020
Publisher : IAES Indonesia Section

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/eecsi.v7.2055

Human-robot interaction can be through several ways, such as through device control, sounds, brain, and body, or hand gesture. There are two main issues: the ability to adapt to extreme settings and the number of frames processed concerning memory capabilities. Although it is necessary to be careful with the selection of the number of frames so as not to burden the memory, this paper proposed identifying hand gesture of video using Spatial Convolutional Neural Networks (CNN). The sequential image's spatial arrangement is extracted from the frames contained in the video so that each frame can be identified as part of one of the hand movements. The research used VGG16, as CNN architecture is concerned with the depth of learning where there are 13 layers of convolution and three layers of identification. Hand gestures can only be identified into four movements, namely 'right', 'left', 'grab', and 'phone'. Hand gesture identification on the video using Spatial CNN with an initial accuracy of 87.97%, then the second training increased to 98.05%. Accuracy was obtained after training using 5600 training data and 1120 test data, and the improvement occurred after manual noise reduction was performed.

Spoken Word and Speaker Recognition Using MFCC and Multiple Recurrent Neural Networks Yoga Utomo; Esmeralda Contessa Djamal; Fikri Nugraha; Faiza Renaldi
Proceeding of the Electrical Engineering Computer Science and Informatics Vol 7, No 1: EECSI 2020
Publisher : IAES Indonesia Section

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/eecsi.v7.2059

Identification of spoken word and speaker has been featured in many kinds of research. The problem or obstacle that persists is in the pronunciation of a particular word. So it is the noise that causes the difficulty of words to be identified. Furthermore, every human has different pronunciation habits and is influenced by several variables, such as amplitude, frequency, tempo, and rhythmic. This study proposed the identification of spoken sounds by using specific word input to determine the patterns of the speaker and spoken using Mel-frequency Cepstrum Coefficients (MFCC) and Multiple Recurrent Neural Networks (RNN). The Mel coefficient of MFCC is used as an input feature for identifying spoken words and speakers using RNN and Long Short Term Memory (LSTM). Multiple RNN works spoken word and speaker in parallel. The results obtained by multiple RNN have an accuracy of 87.74%, while single RNNs have 80.58% using Adam of new data. In order to test our model computational regularly, the experiment tested K-fold Cross-Validation of datasets for spoken and speakers with an average accuracy of 86.07%, which means the model to be able to learn on the dataset without being affected by the order or selection of test data.

Multi-channel of electroencephalogram signal in multivariable brain-computer interface Esmeralda Contessa Djamal; Dimas Andhika Sury
IAES International Journal of Artificial Intelligence (IJ-AI) Vol 12, No 2: June 2023
Publisher : Institute of Advanced Engineering and Science

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/ijai.v12.i2.pp618-626

Brain-computer interface (BCI) usually uses Electroencephalogram (EEG) signals as an intermediate device to drive external devices directly from the brain. The development of BCI capabilities is carried out by involving multivariable EEG signals as movement commands. EEG signals are recorded using multi-channel, enriching information if it uses the suitable method and architecture. This research proposed a two-dimensional convolutional neural networks (CNN) method to recognize multi-channel EEG signals. The vertical dimension is the channel, while the horizontal is the signal sequence. Hence, the signal is connected with the information time series of the same channel and between channels simultaneously. BCI was arranged with multivariable signals, specifically motor imagery and emotion. Both variables have different characteristics, and the information is from different channels. Therefore, it needs multiple CNNs to recognize the two variables in the EEG signal. The experiment showed that the accuracy of multiple 2D-CNN increased to 94.62% compared to 85.44% of single 2D CNN. Multiple 2D-CNN gave accuracy from 82.04% to 94.62% more than multiple 1D-CNN. 2D-CNN makes the channel extraction perfect into vectors to maintain the signal sequence. Signal extraction is essential, so the used Wavelet filter upgraded accuracy from 73.75% to 94.62%.

Title

Found 9 Documents
Search

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Title Search

Found 9 Documents Search

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Title

Found 9 Documents
Search