Data in the health sector are often lacking and unbalanced. It is because collecting data takes time and many resources. One example is sleep apnea data which takes about 8–10 hours to get data and uses specialized hardware like polysomnography (PSG). This study proposes a data augmentation technique to handle unbalanced data using DCGAN and several deep learning models such as 1D-CNN, ANN, LSTM, and 1D-CNN+LSTM as a classifier for apnea detection. The DCGAN architecture used is CNN on the generator and discriminator. DCGAN will create new synthetic data by mimicking the original dataset. This experiment uses a dataset from PhysioNet, the Apnea-ECG, and the MIT-BIH PSG Database. Furthermore, the dataset is preprocessed to remove noise, and the features are extracted manually. The test scenario is to create 10% synthetic data and 50% sleep apnea data to be added to the original dataset. Then compare the performance of multiple deep learning models before and after adding data. The results indicate that augmentation with DCGAN can improve the performance of almost all models, with the highest increase of 1.78% on the 1D-CNN+LSTM model and 4.80% on the LSTM model for the Apnea-ECG and MIT-BIH datasets, respectively.
30 September 2023 in Engineering, MAthematics and Computer Science Journal (EMACS)
This study aims to identify and categorize the BISINDO sign language dataset, primarily consisting of image data. Deep learning techniques are used, with three pre-trained models: ResNet50 for training, MobileNetV4 for validation, and InceptionV3 for testing. The primary objective is to evaluate and compare the performance of each model based on the loss function derived during training. The training success rate provides a rough idea of the ResNet50 model's understanding of the BISINDO dataset, while MobileNetV4 measures validation loss to understand the model's generalization abilities. The InceptionV3-evaluated test loss serves as the ultimate litmus test for the model's performance, evaluating its ability to classify unobserved sign language images. The results of these exhaustive experiments will determine the most effective model and achieve the highest performance in sign language recognition using the BISINDO dataset.
1 January 2024 in Communications in Mathematical Biology and Neuroscience