Korean Journal of Artificial Intelligence, KODISA Foundation

PDF Citaion Performance Evaluation of Bio-Inspired Hybrid GOA-WOA Algorithm for Clustering and Routing in Vehicular Ad-Hoc Networks (VANETs)

Yeonwoo LEE(Mokpo National University) pp.1-8 https://doi.org/10.24225/kjai.2025.13.1.1

초록보기

Abstract

This paper presents a performance evaluation of a novel hybrid bio-inspired algorithm for clustering and routing optimization in Vehicular Ad-Hoc Networks (VANETs). The proposed approach integrates the Grasshopper Optimization Algorithm (GOA) and Whale Optimization Algorithm (WOA) to enhance clustering efficiency, while incorporating established routing protocols such as Ad-hoc On-Demand Distance Vector (AODV) and Greedy Perimeter Stateless Routing (GPSR) to improve overall network performance. The hybrid GOA-WOA clustering algorithm leverages the exploration capabilities of GOA and the exploitation strengths of WOA to achieve optimal cluster formation. This combination aims to balance cluster stability and adaptability in the highly dynamic VANET environment. The clustering results are then utilized by AODV and GPSR routing protocols to establish efficient communication paths between vehicles. Performance is assessed using a Manhattan grid-based mobility model, evaluating metrics like average cluster size, number of clusters, packet delivery ratio, and latency under various node densities and transmission ranges. Simulation results show that this hybrid approach significantly enhances VANET performance compared to traditional methods, particularly in urban scenarios with varying vehicle densities. The proposed hybrid GOA-WOA achieves average 3% improvement in packet delay ratio (PDR) compared to traditional methods. The GOA-WOA clustering improves stability, while AODV and GPSR benefit from optimized cluster structures, reducing routing overhead. The algorithm's adaptive nature maintains optimal performance across different network conditions, demonstrating potential for real-world VANET applications with artificial intelligence (AI) algorithms.

PDF Citaion Enhancing Public Service Satisfaction through Metaverse Integration: Urban Development, Social Interaction, and Policy Implications

Yooncheong CHO(KDI School of Public Policy and Management) pp.9-17 https://doi.org/10.24225/kjai.2025.13.1.9

초록보기

Abstract

Purpose: This study aims to explore how integrating the metaverse can enhance public satisfaction with public services by examining its perceived effects on urban development, social interactions, and promotional policies. Specifically, it addresses the following research questions: 1) How do metaverse-enabled public services that enhance community development and promote the sustainable residential areas influence public satisfaction? 2) How do metaverse-enabled public services fostering social interactions among citizens affect public satisfaction? 3) How do government policies supporting metaverse-based public services impact public satisfaction? Research design, data and methodology: An online survey was conducted in collaboration with a reputable research firm. This study employed multiple regression analysis to test the proposed hypotheses. Results: The findings indicate that metaverse-enabled public services enhancing urban development in residential areas and fostering positive social interactions significantly improve public satisfaction. Furthermore, government policies promoting and revitalizing public services through metaverse technologies have a substantial positive effect on public satisfaction. Conclusions: The results provide valuable managerial and policy insights. They highlight the potential of metaverse technologies as effective tools for city marketing and citizen relationship management, contributing to stronger local communities and enhanced societal connectivity.

PDF Citaion A Study on Driver State Recognition Using CNN-based Multimodal Multi-input Learning

Sooah SHIN(Korean Bible University) ; Jisu KANG(Korean Bible University) ; Sooah KIM(Korean Bible University) ; Hwanseo YEO(Korean Bible University) ; Dongyeon LEE(Korean Bible University) ; Jeonghyun LEE(Korean Bible University) ; Gijun HAN(Korean Bible University) ; Jinho HAN(Korean Bible University) pp.19-25 https://doi.org/10.24225/kjai.2025.13.1.19

초록보기

Abstract

Accurate identification of a driver’s state is extremely important for the safety of both the driver and passengers. To date, many studies have demonstrated that identifying a driver’s state by creating an image of their face and upper body using a camera and then learning these CNN-based neural networks can be effective to a certain extent. In addition, efforts have been made to identify a driver’s level of fatigue using electroencephalogram (EEG), heart rate (ECG), and electrooculogram (EOG) data measurements, multiple sensors, and the RNN-LSTM model method. At present, increasing accuracy can be achieved using simple sensor devices and AI structures that can be installed in small devices such as on-device AI. In this study, we used a driver’s facial expression and upper body along with the sound data generated during driving to recognize the driver’s state more accurately. Subsequent experiments revealed that CNN-based neural network learning alone, using triple input elements, improved accuracy. To apply on-device AI, we proposed a CNN with a simple structure that can collect data employing only a camera and a recorder. We compared the proposed method with learning in ResNet50 and Xception, which revealed that it works effectively. These experimental results indicate that CNN can be used in multimodal applications and can be an efficient choice over other complex neural network learning methods that utilize multimodal learning data.

PDF Citaion Dynamic Subtitle Allocation: A Practical System with Speaker Separation

Hyun-Seok CHOI(Seoul National University) ; Hyuk-Hee HAN(KAIST (Korea Advanced Institute of Science and Technology, Korea).) pp.27-34 https://doi.org/10.24225/kjai.2025.13.1.27

초록보기

Abstract

Subtitles help viewers understand the contents of a video by displaying spoken words or their translations. However, when multiple speakers speak simultaneously, it becomes difficult to determine who is speaking through subtitles alone. Additionally, most subtitles are positioned below the scene, which can disrupt visual focus. To address these issues, we develop a dynamic subtitle allocation pipeline that positions subtitles near the speaker’s face. A speaker’s face is cropped using YOLOv3-based face detection, followed by a lip-reading model associating ground-truth subtitles with the correct speaker. Experimental results show that our method effectively assigns subtitles to the corresponding speakers, even in multi-speaker scenarios, demonstrating its robustness in handling complex audio-visual conditions. This practical approach enhances accessibility, particularly for viewers with hearing impairments, and improves the viewing experience by ensuring seamless subtitle synchronization with speakers. Furthermore, our system operates with standard single-channel audio and requires no additional pre-processing, making it easy to integrate into existing content and deploy in real-world applications.

PDF Citaion Optimal Machine Learning Model Selection for Predicting Cancer Drug Response Using Genomic Data

Geon AN(Eulji University) ; Min Soo KANG(Eulji University) pp.35-41 https://doi.org/10.24225/kjai.2025.13.1.35

초록보기

Abstract

In this paper, we investigate the optimal machine learning model for predicting drug response in cancer cells by leveraging genomic data, with an emphasis on clinical applicability. Utilizing the Cancer Drug Sensitivity Genomics dataset, we integrated diverse genetic characteristics, including gene mutations, copy number variations, and gene expression levels, along with drug response data. A structured data preprocessing pipeline was implemented, including mode replacement for tissue and cancer types, K-Nearest Neighbors imputation for genetic features, and Random Forest Regressor for handling missing numerical values. Regression models, such as Random Forest, K-Nearest Neighbors, Decision Tree, and CatBoost, were trained and evaluated for predictive performance. Experimental results revealed that the CatBoost model outperformed others, achieving a mean squared error of 1.5618, mean absolute error of 0.9355, and an R² score of 0.7855, with the Random Forest model showing comparable performance. These findings highlight the CatBoost model as a robust tool for predicting cancer drug response. Furthermore, this research underscores its potential integration into clinical decision-making systems by enabling personalized drug selection based on patient-specific genetic profiles. Future research may extend this approach to incorporate additional omics data and validate the model's utility in real-world clinical scenarios.

바로가기메뉴