Data Augmentation Effects on Highly Imbalanced EEG Datasets for Automatic Detection of Photoparoxysmal Responses

Martins, Fernando Moncada; Suárez, Víctor Manuel González; Flecha, José Ramón Villar; García-López, Beatriz

doi:10.3390/s23042312

Cited by 13 publications

(5 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Features from statistical, spectral, and temporal domains were extracted for PPR detection. Two neural network models have achieved high accuracy of about 95% for (2C-KNN) and 98% for (DL-NN), demonstrates effective PPR event detection through model dataset balancing and feature extraction techniques [19].…”

Section: Related Workmentioning

confidence: 97%

Early Detection of Stress and Anxiety Based Seizures in Position Data Augmented EEG Signal Using Hybrid Deep Learning Algorithms

Kamakshi,

Rengaraj

2024

IEEE Access

View full text Add to dashboard Cite

Epilepsy is a neurological problem due to aberrant brain activity. Epilepsy diagnose through Electroencephalography (EEG) signal. Human interpretation and analysis of EEG signal for earlier detection of epilepsy is subjected to error. Detection of Epileptic seizures due to stress and anxiety is the major problem. Epileptic seizure signal size, and shape changes from person to person based on their stress and anxiety level. Stress and anxiety based epileptic seizure signals vary in amplitude, width, combination of width and amplitude. In this paper, Seizures of different size and shape are synthesized using data augmentation for different stress and anxiety level. Different augmentation such as (i) position data augmentation (PDA) (ii) random data augmentation (RDA) applied to BONN EEG dataset for synthetizations of stress and anxiety based epileptic seizure signals. Augment EEG epileptic seizure signals are analyzed using proposed methods such as i) FCM-PSO-LSTM and ii) PSO-LSTM for earlier detection of stress and anxiety-based seizures. The proposed algorithms perform better in earlier detection of stress and anxiety-based seizure signals. The predicted accuracy of proposed methods such as i) FCM-PSO-LSTM and ii) PSO-LSTM is about i)98.5% and ii) 97%, for PDA and for RDA accuracy is about i) 98% and ii) 98.5%, respectively.

show abstract

Section: Related Workmentioning

confidence: 97%

Early Detection of Stress and Anxiety Based Seizures in Position Data Augmented EEG Signal Using Hybrid Deep Learning Algorithms

Kamakshi,

Rengaraj

2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…To address dataset imbalance [18], Martins et al [19] in their projects has employed a DA stage in an automatic PPR detection technique. A cross-validation based approach is applied on PPR and non-PPR windows, where non-PPR windows are found under sampled, while to attain the increased number of PPR windows.…”

Section: Class Imbalance Problem In Eeg Datamentioning

confidence: 99%

Ensemble Methods with Statistics and Machine Learning on the Class Imbalance Problems of EEG data

Sneha Mishra

2024

jes

View full text Add to dashboard Cite

Class imbalance in EEG data sets is a significant issue that affects the quality of outputs. The uncertainty in the data sets, which can be small or large, can lead to class imbalance problems (CIP). This imbalance can lead to highly imbalanced predictive models. The selection of random samples for algorithms can result in high variation of classes. Data sets of EEG are generated as image data sets and are often random and never repeat, causing a high variation in classes. To address this issue, various sampling methods on the data and heuristics are being developed to develop predictive models which were in vogue based on the level of the uncertain state.

show abstract

“…This work deals with the MIT imbalanced database where the number of samples belonging to a normal class is more than the number of samples in all reset classes. This problem caused the degradation of the classification model performance in terms of accuracy due to the biased results toward the majority class and ignoring the minority class due to considering it as noise data [30]. On the other hand, many machine learning algorithms are designed with the assumption that the database is balanced [31].…”

Section: Data Preprocessingmentioning

confidence: 99%

Developing Hybrid CNN-GRU Arrhythmia Prediction Models Using Fast Fourier Transform on Imbalanced ECG Datasets

Oleiwi,

AlShemmary,

Al-Augby

2024

MMEP

View full text Add to dashboard Cite

There are many methods to diagnose heart disease; the most effective way is to analyze electrocardiogram (ECG) signals. Generally, the automatic classification techniques based on ECG analysis consist of three steps: data preprocessing, feature extraction, and classification. This study designed eight hybrid model architectures using several types of deep neural networks, including Convolution Neural Network (CNN), Gated Recurrent Unit (GRU), and Bidirectional GRU (Bi-GRU), four of them without Fast Fourier Transform (FFT) and the rest using FFT. Firstly, the MIT-BIH arrhythmia database is cleaned using the wavelet (WT) thresholding method that separates the combined noise and signal frequencies, making it ideal for processing nonstationary ECG signals. Additionally, the imbalance problem in this database was addressed using the synthetic minority over-sampling technique (SMOTE), which is more suitable for medical data than random synthesis methods. Secondly, hybrid models FFT-CNN, FFT-GRU, FFT-CNN-GRU, and FFT-CNN-Bi-GRU are constructed using the new proposed architecture by concatenating resultant features from two paths, the first path using ECG in the time domain and the second path using the resultant spectrum of ECG from FFT as input. A comparative study of the performance of all models was created in terms of accuracy, training time, number of trainable parameters, and robustness against noise. The results show that the proposed CNN, GRU, CNN-GRU, and CNN-Bi-GRU models without WT and FFT achieved 90%, 93%, 95%, and 96% accuracies, while the proposed FFT-CNN, FFT-GRU, FFT-CNN-GRU, and FFT-CNN-Bi-GRU models achieved 97%, 95%, 96%, and 97% accuracies with WT. So, the proposed FFT-CNN model was the best, with less training time and parameters than other models, which significantly impacts designing a high-efficiency model with less complexity for a practical medical diagnosis system. On the other hand, using FFT improved all models' performance, accuracy and robustness against noise.

show abstract

Data Augmentation Effects on Highly Imbalanced EEG Datasets for Automatic Detection of Photoparoxysmal Responses

Cited by 13 publications

References 46 publications

Early Detection of Stress and Anxiety Based Seizures in Position Data Augmented EEG Signal Using Hybrid Deep Learning Algorithms

Early Detection of Stress and Anxiety Based Seizures in Position Data Augmented EEG Signal Using Hybrid Deep Learning Algorithms

Ensemble Methods with Statistics and Machine Learning on the Class Imbalance Problems of EEG data

Developing Hybrid CNN-GRU Arrhythmia Prediction Models Using Fast Fourier Transform on Imbalanced ECG Datasets

Contact Info

Product

Resources

About