International Journal of Computer Science and Humanitarian AI

Two-Dimensional Segmentation to Reconstruct Three-Dimensional Covid-19 Patient’s Lung CT Using Active Contour

2025-01-08T23:11:22+00:00

Beginning in December 2019, SArS-CoV-2, also referred to as COVID-19, quickly spread over the world. With two recurrent waves and a 3.3% fatality rate, COVID-19 has caused over 4 million cases in Indonesia. RT-PCR, antigen, and RT-LAMP are currently the main techniques for COVID-19 detection and diagnosis. A CT scan is usually used for additional diagnosis when RT-PCR results are uncertain, but extra confirmation is required. The need to inform patients about the effects of COVID-19 on the lungs is increasing as the number of cases of the virus keeps rising and diagnosis and first aid techniques advance. The severity of COVID-19-induced pneumonia, which shows up as ground-glass opacities (GGO), which are gray patches in the lung cavity, may be seen on a single-slice CT scan. The degree of lung injury can be measured using image processing techniques. In this study, two- and three-dimensional representations of the lungs were created utilizing a multi-slice CT scan and image processing techniques like active contour and marching cubes. The suggested approach produced an average volume difference of 5% and an accuracy of 62% based on intersection over union (IoU).

Systematic Literature Review of The Use of Music Information Retrieval in Music Genre Classification

2025-02-20T10:12:47+00:00

Emphasizing deep learning models such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), this article explores the application of Music Information Retrieval (MIR) techniques in music genre categorization. These algorithms outperform traditional methods in capturing complex audio patterns, showcasing their potential in advancing music classification tasks. Accurate genre classification critically depends on key features such as spectral, temporal, and timbral characteristics, which play a pivotal role in distinguishing musical styles. However, the performance of these models is heavily influenced by the quality and diversity of the training datasets. Additionally, challenges like model interpretability and reliance on large datasets are addressed. This research utilized a Systematic Literature Review (SLR) to investigate the capabilities of advanced MIR techniques in enhancing music categorization systems, particularly for educational applications and personalized music recommendations. The findings reveal that analyzing the importance of spectral, temporal, and timbral features—key components of MIR—can significantly boost the accuracy and reliability of music genre classification.

Smoker Melanosis Classification Using Oral Photographic Feature Extraction Based On K-Nearest Neighbor

2025-02-20T10:11:24+00:00

Smoking is one of the causes of various diseases in the body. Smoking can also cause abnormal conditions that are pathological and physiological in the oral cavity, one of which is smoker melanosis. The clinical picture of pigmentation smoker melanosis is the presence of scattered brown spots with a diameter of less than 1 cm and is most often located on the gingiva. The data was taken using the oral photograph image capture method using a 12MP resolution camera, provided that the object distance from the camera was 6 cm and the flash was on. This analysis utilized the Gingiva Pigmentation Index (GPI) classification system proposed by Hedin, which compares the pigmented area, and Dummett's Oral Colour Index (DOPI), which assesses the density of pigmentation. In this study, the classification process was carried out with the KNN algorithm using features from digital image processing in the segmentation area, the average value of the red, green, and blue colour levels. The classification process uses the nearest neighbour value of 3 and a p-value of 2 to measure the distance to the nearest neighbor using the Minkowski distance formula. The results of the test data accuracy (1.0) with F1 scores for each class for test data DOPI 0 = 1.0, DOPI 1 = 1.0, DOPI 2 = 1.0, DOPI 3 = 1.0. Meanwhile, the classification process can use more up-to-date methods, such as CNN to improve classification accuracy.

The Framework of Vehicle Detection and Counting System for Handling of Toll Road Congestion using YOLOv8

2025-02-20T02:19:11+00:00

The Global COVID-19 pandemic and the increasing number of vehicles have exacerbated traffic congestion, particularly in developing countries. In Jakarta, Indonesia, congestion on toll roads is a significant issue that needs to be addressed through an Intelligent Transportation System (ITS). One of the key solutions proposed is vehicle detection and traffic prediction on toll roads. This study introduces a computer vision-based approach utilizing YOLOv8 to detect, track, and count vehicles to predict traffic congestion. The system operates by identifying vehicles (cars and trucks), preprocessing the data, and calculating the total number of vehicles within the camera’s range. If the vehicle count surpasses the threshold set by the toll road provider, the system updates the traffic status (normal or congested) and triggers a warning. The vehicle detection system can identify cars and trucks within a range of up to 150 meters. Experimental results using test videos demonstrate that the YOLOv8-based system achieves an accuracy of 98% with an average detection speed of 83.6 milliseconds, ensuring highly efficient performance. With its high accuracy and speed, this system can be effectively integrated into traffic management solutions to alleviate congestion and enhance transportation efficiency in Jakarta.

A Literature Review on AI and DSS for Resilient and Sustainable Humanitarian Logistics

2025-02-20T00:35:50+00:00

Disaster response is a critical component of disaster management, requiring effective strategies to reduce exposure and vulnerability to hazards. Rising global temperatures and extreme weather events have intensified the need for adaptive disaster relief systems. Humanitarian logistics, a vital subset of the supply chain, plays a central role in disaster preparedness, response, and recovery phases but often faces challenges such as resource constraints, inefficient communication, and unpredictable crises. This study employs a systematic literature review (SLR) using the PRISMA methodology to explore the application of Artificial Intelligence (AI) and Decision Support Systems (DSS) in humanitarian logistics from 2019 to 2024. SCOPUS served as the primary database, identifying 1,171 documents, with 52 studies selected for in-depth analysis. These studies highlight the potential of AI techniques, including machine learning and clustering algorithms, and DSS implementations for resource allocation, stakeholder coordination, and real-time decision- making. Findings demonstrate that integrating AI and DSS can optimize emergency vehicle routing, improve relief distribution, and enhance stakeholder collaboration. Advanced technologies such as Radio Frequency Identification (RFID), the Internet of Things (IoT), and Digital Twins improve logistics efficiency and scalability. Despite these advancements, challenges like data integration and algorithmic reliability persist. The study recommends prioritizing transparent systems, hybrid simulations, and addressing algorithmic constraints to advance disaster management practices.

Comparison of Machine Learning Classification Models in Predicting The Titanic Survival Rate

2025-01-06T03:55:06+00:00

The tragic sinking of the Titanic in 1912 has been a subject of great interest, particularly in analyzing the factors that influenced passenger survival rates. This study applies machine learning techniques to predict the survival of Titanic passengers based on various attributes. The dataset used includes demographic details and passenger-specific features such as age, gender, ticket class, number of siblings/spouses, number of parents/children traveling, ticket fare, and departure location. An exploratory data analysis is conducted to understand patterns within the dataset, followed by data preprocessing steps, including handling missing values and encoding categorical variables. To develop the predictive model, multiple machine learning algorithms are implemented, including Logistic Regression, Random Forest, Extra Trees, Decision Tree, LGBM Classifier, and XGBoost Classifier. The results indicate that the Random Forest model achieves the highest accuracy at 0.815, while the LGBM Classifier attains the highest cross-validation score of 0.821. Feature importance analysis highlights gender and ticket class as the most significant factors affecting survival probability. This study demonstrates the effectiveness of machine learning classification techniques in analyzing historical data and predicting binary outcomes. The insights gained from this research can be applied to other domains involving historical data analysis and classification tasks, such as risk assessment, medical prognosis, and social science research. By leveraging machine learning, this approach provides a data-driven perspective on historical events, enabling better decision-making in similar predictive modeling scenarios.

Editorial, Foreword, and Table of Content

2025-02-20T20:08:42+00:00