Jurnal Elektro

Deep Learning-Based Brain Tumor Classification Using Convolutional Neural Network

Naumi Sasita, Futri Zalzabilah Ray, M.Fauzanil Wildan A.R., Tantowi Hutagalung, Rokhmat Febrianto

Abstract:

An essential noninvasive medical diagnostic technique is magnetic resonance imaging (MRI), which is particularly useful for identifying brain cancers. While earlier algorithms proved effective on smaller MRI datasets, their performance suffered on bigger datasets. This study addresses the need for a swift and reliable brain tumor classification system capable of sustaining optimal performance across comprehensive MRI datasets. The convolutional neural network is implemented using the Keras library, incorporating the ResNet50 architecture as a pre-trained model. The ResNet50 model is fine-tuned for the specific brain tumor classification task, with a Global Average Pooling layer, dropout, and a final dense layer with softmax activation. Data augmentation techniques are employed to enhance the model’s robustness, including rotation, width and height shifts, and horizontal flips. The training process involves optimizing the model using the Adam optimizer with a learning rate of 0.0001. Early stopping, learning rate reduction on plateau, and model checkpointing are implemented as callbacks to ensure efficient training and prevent overfitting. The proposed model achieves a remarkable accuracy of 99.47 percent after 15 epochs. The classification task involves distinguishing among four classes: glioma, meningioma, pituitary, and no tumor.

Classification Of Multi-Class Face Expression Using Modification Of VGG-16 Model

Aryadana Priyatama, Sugiyanto

Abstract:

In the era of modern technology, facial recognition has become an important application in various fields, such as security, education and health. One method used to recognize faces is a Convolutional Neural Network (CNN), specifically the VGG-16 architecture which is known for its consistent performance. But even though CNN can recognize faces, its accuracy in recognizing faces is inadequate. This research aims to increase the accuracy of facial expression classification so that it is more optimal by modifying the CNN VGG-16 architecture. This research uses GridSearch techniques, K-Fold Cross Validation, and utilizes multiple datasets. The dataset used consists of two image datasets, namely SMIC and SAMM facial-micro expressions, each of which has been normalized and converted to a grayscale scale measuring 48x48 pixels. The GridSearch process is applied to optimize parameters such as the number of filters, learning rate, dropout rate, activation function, and batch size. The K-Fold Cross Validation technique with five folds was used to ensure the generalization of the model to new data. The research results show that this modification is able to achieve validation accuracy of up to 98.31% in the training process, showing a significant improvement compared to the standard method. And showed an increase in accuracy in testing of 98.04% in research.

Performance Analysis of Deep Neural Network to Noisy Digit Dataset

Karel Octavianus Bachri

Abstract:

This work investigates the impact of noise on model performance by training a neural network on a digit dataset with varying Signal-to-Noise Ratios (SNR) to assess its resilience and generalization ability. The experimental setup involved training the model on datasets with noise levels ranging from clean images to highly distorted ones (SNR 5%–25%), analyzing accuracy, mini-batch loss, and training time. Results indicate that while the model achieves high accuracy (96.88%) at mild noise levels (SNR 5%), performance declines significantly at higher noise levels, with accuracy dropping to 78.91% at SNR 25%. The analysis of mini-batch loss and training time reveals that noise slows convergence and increases computational complexity. The confusion matrix further confirms that while the model effectively distinguishes between classes, noise-induced misclassifications become more frequent at lower SNRs. These findings emphasize the importance of noise reduction techniques and data preprocessing to improve model robustness in real-world applications.

PENGUKURAN BACKUP TIME UNINTERRUPTIBLE POWER SUPPLY UNTUK PENENTUAN KAPASITAS BATERAI VRLA

Jong Susanto, Melisa Mulyadi, Linda Wijayanti

Abstract:

Baterai adalah perangkat yang menyimpan daya di dalam sistem Uninterruptible Power Supply (UPS). Jenis baterai UPS yang paling umum dijumpai saat ini adalah Valve Regulated Lead Acid (VRLA). Agar UPS dapat bekerja secara maksimal sesuai dengan kapasitas dan backup time nya, maka perlu menentukan kapasitas baterai. Pada penelitian ini dilakukan penentuan kapasitas baterai VRLA pada UPS tertentu berdasarkan pengujian langsung dan berdasarkan perhitungan. Perhitungan kapasitas baterai VRLA mengacu pada datasheet UPS dan baterai. Tahapan perhitungan dimulai dari  menghitung arus dan daya baterai yang digunakan sesuai dengan daya beban yang diserap UPS serta  mengacu pada tabel karakteristik constant current discharge dan constant power discharge. Penentuan kapasitas berdasarkan pengujian dilakukan dengan mengukur backup time dari UPS tersebut. Hasil perhitungan dan hasil pengujian langsung menunjukkan adanya perbedaan backup time karena baterai yang digunakan adalah baterai stock lama. Hasil penelitian juga membuktikan bahwa penentuan kapasitas baterai dengan perhitungan dapat memastikan kapasitas baterai yang lebih sesuai atau tidak terlalu besar sehingga lebih ekonomis.