Public Research Journal of Engineering, Data Technology and Computer Science

Applying A Supervised Model for Diabetes Type 2 Risk Level Classification

2024-02-02T04:21:15+00:00

Diabetes can lead to heart attacks, kidney failure, blindness, and increased risk of death. This research was conducted with the aim of classifying a diabetes risk dataset. In this context, performance comparison was carried out on three supervised learning algorithms: K-Nearest Neighbor, Naive Bayes, and Random Forest, against a dataset containing information on specific indicators related to diabetes risk. Additionally, this study also aimed to evaluate the accuracy comparison of the results produced by these three algorithms. The results of this research show that Random Forest performs very well in detecting diabetes, prediabetes, and non-diabetes, with high precision, recall, and F1-score levels. Meanwhile, although the results are still below Random Forest, both Naive Bayes and K-NN still demonstrate significant performance, especially regarding prediabetes cases. In conclusion, from the comparison results, the Random Forest algorithm shows the highest accuracy level at 99%, followed by K-Nearest Neighbor with an accuracy of 85%, while Naive Bayes has the lowest accuracy rate of 74%. This research indicates that the Random Forest algorithm excels in classifying data compared to the other two algorithms.

Implementation of Gated Recurrent Unit, Long Short-Term Memory and Derivatives for Gold Price Prediction

2024-08-24T07:27:58+00:00

Gold is a precious metal with high resale value, often considered a safe investment as its price typically rises with inflation, attracting investors. However, even slight changes in gold prices can have significant impacts. To build an accurate forecasting model, this study applies and compares Gated Recurrent Unit (GRU), Bidirectional Gated Recurrent Unit (Bi-GRU), Long Short-Term Memory (LSTM), and Bidirectional Long Short-Term Memory (Bi-LSTM) algorithms on global gold prices. GRU and LSTM are recurrent neural networks designed to capture patterns in sequential data, where GRU uses a simplified gating mechanism to retain essential information, and LSTM, with its more complex gates, helps manage long-term dependencies in data. Bi-GRU and Bi-LSTM process data bidirectionally, capturing context from both past and future sequences for better prediction accuracy. This research uses data from Yahoo Finance (01-01-2014 to 12-06-2024) and experiments with optimization techniques (Adam, AdamW, Adamax, and Nadam), batch sizes (8, 16, and 32), time steps (10, 20, and 30), and a learning rate of 0.0001, trained for 1000 epochs with checkpoints and early stopping. Bi-GRU with Nadam, batch size 8, and 20 time steps proved most effective, with MSE of 4.1153, RMSE of 2.0286, MAE of 1.5881, and MAPE of 0.8857%. Forecasts using this model predict a 20-day decline in gold prices.

Comparative Analysis of Weather Image Classification Using CNN Algorithm with InceptionV3, DenseNet169 and NASNetMobile Architecture Models

2024-08-24T07:21:38+00:00

Rapid weather changes have a significant impact on various aspects of human life, including social and economic development. Weather analysis traditionally relies on data from Doppler radar, weather satellites, and weather balloons. However, advancements in computer vision technology provide new opportunities to enhance weather prediction systems through image recognition and classification. Studies evaluating and comparing deep learning architectures for weather image classification remain limited.This research utilizes Convolutional Neural Networks (CNN) to classify weather images using three architectures: InceptionV3, DenseNet169, and NASNetMobile. The results show that InceptionV3 achieved 97.94% accuracy on training data, 92.34% on validation data, and 93.81% on test data. DenseNet169 achieved 98.09% accuracy on training data, 88.46% on validation data, and 92.33% on test data. NASNetMobile achieved 96.51% accuracy on training data, 87.82% on validation data, and 89.97% on test data. Based on these results, InceptionV3 is the optimal choice for weather classification due to its consistent performance.This research addresses the gap in evaluating CNN architectures for weather data and contributes to improving weather monitoring systems, early disaster warnings, and applications reliant on accurate predictions. These findings also provide a foundation for the development of advanced technologies in image analysis and weather forecasting in the future.

A Deep Learning Approach Bassed on Classification to Detect Facial Skin Defect

2024-11-13T11:29:08+00:00

As people are more active, facial skin is often neglected, which can lead to acne, eye bags, and redness. In this study, deep learning models such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Generative Adversarial Networks (GANs) are used to classify facial skin damage. DenseNet201 and MobileNetV2 architectures were also used to evaluate the models in this study. The dataset used consists of facial skin disease photos collected from the Kaggle database. The model was trained and tested to classify the types of skin damage after going through data collection and preprocessing stages. The results showed that the GANs model and the DenseNet201 and MobileNetV2 architectures were the best models, with test accuracy values of 89% for the GANs model, 88% for the DenseNet201 architecture, and 89% for the MobileNetV2 architecture. These results show that deep learning approach techniques can help classify and find facial skin problems well. and it is expected that it will be a great progress in the field of dermatology and skin health.

Sentiment Analysis of Twitter Reviews on Google Play Store Using a Combination of Convolutional Neural Network and Long Short-Term Memory Algorithms

2024-09-12T03:09:46+00:00

In this era of rapidly evolving technology, the use of social media has become widespread and has become a major platform for sharinhabibahdian.khalifah@ogr.deu.edu.trg people's opinions and views. Google Play Store, as one of the main platforms for digital content, provides access to various applications including Twitter, which allows users to provide reviews and ratings. This research aims to conduct sentiment analysis of Twitter reviews on the Google Play Store using two algorithms namely Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). The data used is 4999 reviews after the scraping process. From the experimental results, an accuracy value of 84.67%, recall of 81%, and precision of 84% were obtained on CNN, and an accuracy of 82.19% recall of 69%, and precision of 87% on LSTM. From these results, it can be seen that the highabibahdian.khalifah@ogr.deu.edu.trhest accuracy value is obtained in the CNN algorithm. Although the difference in accuracy is small, the CNN algorithm provides better results in classifying sentiment analysis data on Twitter reviews on the Google Play Store.

Analysis of Zoom App User Reviews on Google Play Store Using Recurrent Neural Networks and Gated Recurrent Unit Algorithms

2024-09-12T03:16:20+00:00

The World Health Organization (WHO) declared COVID-19 a global pandemic on March 11, 2020. Technology is crucial to stop the spread of the virus. Video conferencing applications such as Zoom Cloud Meetings are essential for collaboration and communication as the government issues policies to conduct various activities from home. Zoom was released in January 2013 to become a trendy video conferencing platform until now. However, post-pandemic, the Zoom App faces challenges maintaining user satisfaction due to the reduced need for virtual meetings. This research aims to analyze user reviews of the Zoom app on the Google Play Store using the RNN and Analysis of Zoom App User Reviews on Google Play Store Using Recurrent Neural Networks (RNN) and Gated Recurrent Unit Algorithms (GRU) algorithms, determine which user reviews are positive, negative, and neutral, identify common problems with Zoom for improvement recommendations, and compare the accuracy between the RNN and GRU algorithms. The results showed that out of 5000 reviews, 3728 sentiments were Positive, 1041 sentiments were Negative, and 231 sentiments were Neutral. The RNN algorithm achieved 86% accuracy, 86% precision, 100% recall, and 92% f1-score, while GRU achieved 83% accuracy, 87% precision, 92% recall, and 89% f1-score. Thus, RNN is superior in sentiment classification and most users are satisfied with the app, but negative reviews indicate areas that require improvement. This research provides valuable insights for developers to improve Zoom app features based on user feedback.