Jurnal Bumigora Information Technology (BITe)

KosyFinder E-Commerce System Based on Web Geographic Information System and Expert System

Hardiana Hardiana — 2024-11-09

Background : The current problem of students and the community in searching for boarding house information only relies on information from friends, relatives or manual searches in searching for boarding house availability. Currently, the availability of information service provider platforms for boarding houses is now abundant, especially those based on websites or android-based. Boarding house information services based on websites or android-based only provide general information related to boarding house information, have not been able to provide comprehensive information.

Objective : The purpose of this study is the KosyFinder e-commerce platform integrated with the Web Geographic Information System (WebGis) and Expert System which can help users in choosing a boarding as needed.

Method : The research method used is Research and Development using the waterfall development model and the certainty factor (CF) method in diagnosing the health of the residential environment around the boarding house.

Result : The results of this study were able to produce an e-commerce application product KosyFinder that was able to be integrated with WebGis and expert systems. The results of expert testing and user testing obtained a final value of 3.8 (very suitable), which means that the resulting system can be used.

Conclusion : The platform that develops is functioning properly and as expected in accordance with the results of expert and user testing in the category is very appropriate.

Classification of Atopic Dermatitis and Psoriasis Skin Diseases Using Residual Network (ResNet-50)

Rakhimatulfitria Mekacahyani — 2024-11-09

Background: Atopic dermatitis and psoriasis are common skin diseases with similar symptoms, characterized by abnormally red or inflamed epidermal lesions and varying degrees of skin thickening. However, they are distinct conditions, making it crucial to understand how to differentiate between them. This understanding can help reduce stigma and the risk of comorbidities, thereby improving patients' quality of life and preventing more serious health risks.

Objective: The aim of this research is to increase accuracy in classifying the skin diseases atopic dermatitis and psoriasis using the Residual Network (ResNet-50) model without overfitting, and compare it with the MobileNet model to find the best approach.

Method: The method used in this study is the ResNet-50 architecture for skin disease classification, namely atopic dermatitis and psoriasis. The selection of the ResNet-50 model is based on the use of shortcut connections that allow the application of deeper networks without experiencing the problem of vanishing gradients.

Result: The results showed that the best accuracy reached 92.75% for training data and 88.00% for testing data, with a data ratio of 80%:10%:10%. In addition, the confusion matrix results from the best model showed that the precision, recall, and F1 score values for both diseases were between ≥80% and ≤96%.

Conclusion: The ResNet-50 method in scenario 1 outperformed other scenarios, improving classification accuracy and enhancing diagnostic effectiveness and medical practice development.

Improving Large Language Model’s Ability to Find the Words Relationship

Sirojul Alam — 2024-11-09

Background: It is still possible to enhance the capabilities of popular and widely used large language models (LLMs) such as Generative Pre-trained Transformer (GPT). Using the Retrieval-Augmented Generation (RAG) architecture is one method of achieving enhancement. This architectural approach incorporates outside data into the model to improve LLM capabilities.

Objective: The aim of this research is to prove that the RAG can help LLMs respond with greater precision and rationale.

Method: The method used in this work is utilizing Huggingface Application Programming Interface (API) for word embedding, store and find the relationship of the words.

Result: The results show how well RAG performs, as the attractively rendered graph makes clear. The knowledge that has been obtained is logical and understandable, such as the word Logistic Regression that related to accuracy, F1 score, and defined as a simple and the best model compared to Naïve Bayes and Support Vector Machine (SVM) model.

Conclusion: The conclusion is RAG helps LLMs to improve its capability well.

Detection of Lumpy Disease in Livestock Using the MobileNetV2 Architecture Method

Dion Pratama Putra — 2024-11-09

Background: Lumpy Skin Disease (LSD) causes skin lesions, decreased milk production, and death in livestock such as cows.

Objective: The purpose of this study is to detect LSD disease quickly and accurately using the Convolutional Neural Network (CNN) MobileNetV2 method based on android application.

Method: This study uses a quantitative method with a reuse-oriented development approach and the MobileNetV2 algorithm trained with augmentation data and LSD disease image classification.

Result: The results of this study are that the MobileNetV2 classification model is able to detect LSD with an accuracy of 95.91%. The developed application makes it easier for farmers to detect diseases early so that they can accelerate preventive measures.

Conclusion: The implications of this study indicate that the MobileNetV2 model can improve the effectiveness of disease detection in livestock and can be applied in animal health applications in the field.

The Lung Sound Classification Using Mel Frequency Cepstral Coefficients and Convolutional Neural Networks

Sayyidis Syariful Halim — 2024-11-22

Background: Challenges in diagnosing respiratory disorders are often caused by the lack of technological tools capable of accurately recognizing lung sound patterns, thereby reducing the potential for subjective misdiagnosis by medical personnel.

Objective: This study aims to develop a lung sound classification model that is able to detect respiratory disorders early and accurately.

Methods: The method used includes a combination of data augmentation techniques and Mel Frequency Cepstral Coefficient (MFCC) feature extraction to improve the performance of Convolutional Neural Network (CNN) in classifying lung sounds. A total of 1,350 lung audio recordings were categorized into nine classes, including normal and abnormal sounds. The augmentation techniques applied include the addition of white noise, pitch scaling, time stretching, and random gain to enrich the variety of training data.

Result: The results show that the E-CNN2D model is able to achieve an accuracy of up to 95%, surpassing the previous model, which had an accuracy range of 83-93%.

Conclusion: With these results, this study has the potential to be a fast and accurate diagnostic tool solution so that it can support medical personnel in reducing the risk of subjective misdiagnosis in respiratory disorders.

Web-Based Pistol Storage Security Monitoring System Optimization for Database Effectiveness

Ahmad Lukman Nur Rokhim — 2024-11-23

Background: Gun storage security is an important aspect in supporting operations, but manual monitoring often creates obstacles to the effectiveness and accuracy of gun storage data management.

Objective: The purpose of this study is to optimize the gun storage security monitoring system with web-based technology to improve the effectiveness and accuracy of the database.

Methods: The method used in this study is the development of a web server-based system that is integrated with real-time monitoring features, automatic notifications, cloud-based data storage to facilitate data access and tracking.

Result: The results of this study are the creation of a system that is able to improve the accuracy and responsiveness of gun storage monitoring, as well as improve the efficiency of database management.

Conclusion: In conclusion, the implementation of this web-based system has the potential to reduce security risks and significantly increase the effectiveness of monitoring in gun storage.