Public Research Journal of Engineering, Data Technology and Computer Science

Amazon Stock Price Prediction Using Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU)

Muthia Tshamaroh — 2025-07-06

Stocks have become one of the largest and most intricate financial markets globally due to their high popularity, making them very challenging to predict as they can process millions of transactions rapidly. The objective of this study is to enhance the field by creating a dependable and accurate model for predicting the stock price of Amazon. This will be achieved via the use of advanced algorithms such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). This research utilised historical data on Amazon's stock price from the past five years, which was acquired from Yahoo Finance. The data was partitioned using a hold-out validation technique, allocating 80% for training and 20% for testing. The model underwent training using different optimizers (Adam, SGD, RMSprop), batch sizes (8, 16, 32), and learning rates (0.001, 0.0001). The evaluation criteria comprised of mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The results suggest that the GRU model, when trained with the RMSprop optimizer using a batch size of 16 and a learning rate of 0.0001, as well as with the SGD optimizer using a batch size of either 16 or 32 and a learning rate of either 0.001 or 0.0001, produced the lowest error metrics, indicating superior performance. This study enables more precise forecasts of stock prices and more efficient investment techniques.

Deep Learning for Pneumonia Detection in Chest X-Rays using Different Algorithms and Transfer Learning Architectures

Danur Lestari — 2025-07-06

Pneumonia is one of the lung conditions brought on by bacterial infections. An accurate diagnosis is necessary for successful treatment. A radiologist can typically diagnose the condition based on images from a chest X-ray. The diagnosis may be arbitrary for a variety of reasons, such as the indistinctness of certain diseases on chest X-ray pictures or the possibility of the illness being mistaken for another. Consequently, clinicians require guidance from computer-aided diagnosis tools. We diagnosed pneumonia using two algorithms CNN and GAN, as well as two architectures ResNet50V2 and InceptionV3. The test results show that the ResNet50V2 architecture is superior to the InceptionV3 architecture on the CNN algorithm with an accuracy of 94% versus 93%. In addition, the test results on the GANs algorithm show that the ResNet50V2 architecture is superior to the InceptionV3 architecture with an accuracy of 96%, while the InceptionV3 architecture achieves an accuracy of 92%.

Leveraging Machine Learning for Early Risk Prediction in Cirrhosis Outcome Patients

Yasir Hussein Shakir — 2025-07-06

Millions of individuals worldwide suffer from liver cirrhosis, which is one of the primary causes of mortality. Healthcare professionals may have more opportunities to treat cirrhosis patients effectively if early death prediction is made and it is postulated that death in this cohort would be correlated with laboratory test findings and other relevant diagnoses. In this study five machine learning models, including LR, SVM, XGBoost, AdaBoost and KNN, are implemented and evaluated. The preprocessing steps included feature selection, categorical data encoding, and data balancing using SVMSMOTE. The XGBoost model demonstrated superior performance, achieving 89.55% accuracy, 89.69% precision, 89.55% recall, and an F1-score of 89.59% after balancing. These findings highlight the potential of machine learning models in accurate risk detection in patients with cirrhosis and providing valuable support in clinical decision-making and improving patient treatment.

Implementation of Deep Learning for Brain Tumor Classification from Magnetic Resonance Imaging

Nur Alfa Husna — 2025-07-06

Brain tumours are a medical problem that causes many people to die in the world due to brain cancer. Brain tumours are one of the dangerous types of brain cancer. MRI is well proven in the assessment of brain tumours, although conventional imaging has limitations in evaluating the extent of the tumour. In the field of medicine, there has been an increase in large amounts of data and traditional models cannot manage such data efficiently. So there is a need for medical image analysis that can store and analyse large medical data efficiently. This research will adopt a deep understanding transfer learning approach with four models namely VGG16, VGG19, MobileNetV2 and ResNet50 to classify 2 types of image shapes that detect whether a person has a brain tumour or not using Magnetic Resonance Imaging (MRI) data with Convolution Neutral Network (CNN). The number of datasets used is 4600 MRI images with 2 classes namely Brain Tumour and Health. The hyperparameters used are image size 224x224 pixels, training data ratio 70%, test data 30%, using Adam optimizer, learning rate 0.0001, using batch size 64 and epoch value 50. The best results in this study were obtained by MobileNetV2 architecture with an accuracy of 88.77%.

Analysis Comparison Classification Image Disease Eye Using the CNN Algorithm, Inception V3, DenseNet 121 and MobileNet V2 Architecture Models

Nasya Amirah Melyani — 2025-07-06

Eye disease is a significant global health problem, with more than two billion people experiencing vision impairment. Some of the main causes of visual impairment include cataracts, glaucoma, diabetic retinopathy, and age-related macular degeneration. Early detection of eye disease is very important to prevent blindness. The fundus of the eye, which includes the retina and blood vessels, is an important area in the diagnosis of retinal diseases. Fundus disease can cause significant vision loss and is one of the leading causes of blindness. Automated analysis of fundus images is used to diagnose common retinal diseases, ranging from easily treatable to very complex conditions. This research discusses eye disease image classification using several Convolutional Neural Network (CNN) architectures, namely Inception V3, DenseNet 121, and MobileNet V2. The dataset used is 4217 fundus images categorized based on the patient's health condition. Data is processed through normalization and augmentation to improve model performance. Experimental results show that MobileNet V2 has the highest accuracy of 81.3%, followed by Inception V3 with 77.3%, and DenseNet 121 with 76.7%. The use of appropriate CNN models in the classification of eye fundus images can help in early detection of eye diseases, thereby preventing further visual impairment.

Classification of E-Commerce Shipping Timeliness Using Supervised Learning Algorithm

Novrian Pratama — 2025-07-06

Developments in the e-commerce sector have increased rapidly since the onset of COVID-19, which has changed consumers' shopping habits. The growth in the number of e-commerce consumers affects the demand for long-distance delivery of goods. The problem of late delivery of goods is one of the challenges that is often experienced, and this can affect the level of customer satisfaction. This study aims to analyze whether the delivery of goods has been carried out according to schedule or has experienced delays. By using e-commerce shipping datasets obtained through the website, this research applies five supervised learning algorithms in the classification process, namely Decision Tree, Naïve Bayes Classifier, K-Nearest Neighbors (K-NN), Random Forest, and Support Vector Machine (SVM). The evaluation results show that dataset sharing using the K-Fold Cross Validation technique provides the best performance at K=8. Support Vector Machine showed the highest level of accuracy of 66.35%, followed by precision of 69.31% and recall of 66.35%. In contrast, the Naïve Bayes Classifier algorithm recorded the lowest performance with accuracy 64.22%, 97.73% precision, and 42.67% recall. These results show that the SVM algorithm is better at classifying the timeliness of delivery compared to the other four algorithms.

Lung Disease Risk Prediction Using Machine Learning Algorithms

Ananda Putri Aulia — 2025-07-06

Lung diseases, including lung cancer, are one of the leading causes of death in the world. Early detection is essential to increase patients' chances of recovery and reduce healthcare costs. The utilization of machine learning algorithms can be used to solve this problem. This study evaluates five machine learning algorithms, namely K-Nearest Neighbors (K-NN), Naïve Bayes Classifier (NBC), Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM), for lung disease prediction using a dataset of 30,000 data with 11 attributes from Kaggle. The dataset was processed through data preprocessing and divided into training and test data with a ratio of 70%:30% and 80%:20%. The algorithm performance was evaluated using precision, recall, F1-score, and accuracy metrics. The results show that RF, SVM, and DT algorithms have the highest performance, with accuracy reaching 94.72% at 70%:30% ratio. The DT algorithm, which previously showed low performance in heart disease classification, provided competitive results in lung disease prediction. This research focuses on the importance of proper algorithm selection and data organization to improve the effectiveness of disease prediction. The findings contribute to the development of artificial intelligence technology for medical applications, particularly in supporting early diagnosis of lung diseases.