Jurnal Online Informatika
Not a member yet
276 research outputs found
Sort by
Performance Evaluation of Vehicular Ad Hoc Networks Considering Malicious Node Impact on Quality of Services Metrics
Full text linkVehicular Ad Hoc Networks (VANETs), a subset of mobile ad hoc networks (MANETs), is essential for enabling communication between vehicles in intelligent transportation systems. However, their dynamic and decentralized nature exposes them to significant security threats, particularly from malicious nodes. Attacks such as black holes and wormholes can severely degrade network performance by causing packet loss and increasing end-to-end delays. This paper aims to evaluate the impact of malicious node behavior on VANET performance using key Quality of Service (QoS) parameters, including throughput, end-to-end delay, jitter, packet delivery ratio (PDR), and packet loss ratio (PLR). The specific objective is to analyze how black hole and wormhole attacks affect communication efficiency in VANET environments. The main contribution of this work lies in the integration of Simulation of Urban Mobility (SUMO) for realistic traffic scenario generation with Network Simulator 3 (NS-3) for detailed network performance evaluation. This approach enables comprehensive simulation of VANET behavior under attack conditions. The findings provide valuable insights into the vulnerabilities of VANETs and form a basis for the design of more robust and secure vehicular communication systems
Optimizing Machine Learning Models for Graduation on Time Prediction: A Comparative Study with Resampling and Hyperparameter Tuning
Full text linkTimely graduation prediction is a crucial issue in higher education, especially when academic, demographic, and behavioral factors interact in complex ways. However, many previous studies rely on default machine learning (ML) parameters and fail to consider the class imbalance problem, leading to suboptimal predictions. This study aims to build a comprehensive framework to evaluate the effectiveness of seven ML algorithms, which are AdaBoost, K-Nearest Neighbors, Naïve Bayes, Neural Network, Random Forest, SVM-RBF, and XGBoost, for predicting graduation on time by incorporating five resampling techniques and hyperparameter tuning. Resampling methods include Random Undersampling (RUS), Random Oversampling (ROS), SMOTENC, and two hybrid approaches (RUS-ROS and SMOTENC-RUS). Hyperparameter tuning was conducted using Grid Search, and model performance was evaluated through cross-validation and hold-out methods. The results show that Random Forest combined with RUS-ROS achieved the best performance, with an average metric score of 0.948. Statistical analysis using PERMANOVA (p = 0.009) and Bonferroni\u27s post-hoc pairwise tests confirmed significant differences between certain models. This study contributes to the educational data mining literature by demonstrating that combining resampling and hyperparameter tuning improves classification performance in imbalanced educational datasets
Application of Self-Organizing Map and K-Means to Cluster Bandwidth Usage Patterns in Campus Environment
Full text linkUnequal bandwidth distribution in campus environments often stems from a lack of understanding of WiFi usage patterns, as seen at Itenas Bandung. Here, bandwidth is allocated equally across all buildings, ignoring differences in demand, leading to inefficiencies in high-usage areas and poor money management due to unnecessary allocation of resources to low-demand buildings. This study aims to optimize bandwidth allocation by analyzing usage patterns using a combination of Self-Organizing Map (SOM) and K-Means clustering methods. SOM is used to group buildings into low, medium, and high bandwidth usage categories, while K-Means refines these clusters to enhance accuracy. The proposed approach demonstrated significant improvements in clustering quality, with the Silhouette Index increasing from 0.321 to 0.773 and the Davies-Bouldin Index dropping from 0.896 to 0.623 in the first test. Similar enhancements were observed in subsequent tests, highlighting the effectiveness of this method in addressing unequal bandwidth distribution. This research offers a practical solution for more efficient network and financial management in educational institutions
Multi-Platform Detection of Melon Leaf Abnormalities Using AVGHEQ and YOLOv7
Full text linkThis research develops a multiplatform system for detecting abnormalities in melon leaves, integrating an Internet of Things (IoT) approach using Jetson Nano, a Streamlit-based website, and a mobile application for real-time monitoring. The system employs preprocessing with Average Histogram Equalization (AVGHEQ) to enhance image quality, followed by modeling with the YOLOv7 algorithm on a dataset of 469 training images and 52 test images, validated through 5-fold cross-validation. The model achieved a mean Average Precision (mAP) of 84% with an inference detection time of 4.5 milliseconds. Implementation on Jetson Nano resulted in a 25% increase in CPU usage (from 25% to 50%) and a 20% increase in RAM usage (from 70% to 90%). By combining these platforms and leveraging robust data preprocessing and modeling techniques, the system provides an accessible, efficient, and scalable solution for agricultural monitoring, enabling farmers to address plant health issues promptly and effectively
Study of the Application of Text Augmentation with Paraphrasing to Overcome Imbalanced Data in Indonesian Text Classification
Full text linkData imbalance in text classification often leads to poor recognition of minority classes, as classifiers tend to favor majority categories. This study addresses the data imbalance issue in Indonesian text classification by proposing a novel text augmentation approach using fine-tuned pre-trained models: IndoGPT2, IndoBART-v2, and mBART50. Unlike back-translation, which struggles with informal text, text augmentation using pre-trained models significantly improves the F1 score of minority labels, with fine-tuned mBART50 outperforming back translation and other models by balancing semantic preservation and lexical diversity. However, the approach faces limitations, including the risk of overfitting due to synthetic text\u27s lack of natural variations, restricted generalizability from reliance on datasets such as ParaCotta, and the high computational costs associated with fine-tuning large models like mBART50. Future research should explore hybrid methods that integrate synthetic and real-world data to enhance text quality and diversity, as well as develop smaller, more efficient models to reduce computational demands. The findings underscore the potential of pre-trained models for text augmentation while emphasizing the importance of considering dataset characteristics, language style, and augmentation volume to achieve optimal results
Road Damage Detection Using YOLOv7 with Cluster Weighted Distance-IoU NMS
Full text linkRoad damage can occur everywhere. Potholes are one of the most common types of road damage. Previous research that used images as input for pothole detection used the Faster Regional Convolutional Neural Network (R-CNN) method. It has a large inference time because it is a two-stage detection method. The object detection method requires post-processing for its detection results to save only the best prediction from the method, namely, non-maximum suppression (NMS). However, the original NMS could not properly detect small, far, and two objects close to each other. Therefore, this research uses the YoloV7 method as the object detection method because it has better mean Average Precision (mAP) results and a lower inference time than other object detection methods; with an improved NMS method, namely Cluster Weighted Distance Intersection over Union (DIoU) NMS (CWD-NMS), to solve small or close potholes. When training YoloV7, we combined a new, independently collected pothole dataset, with previous public research datasets, where the detection results of the YoloV7 method were better than those of Faster R-CNN. The YoloV7 method was trained using various scenarios. The best scenario during training is using the best checkpoint without using a scheduler. The mAP.5 and mAP.5-.95 value of CWD-NMS was 89.20% and 63.30% with 10.30 millisecond per image for inference time
A Trust-Based Reputation System for Security in the Internet of Vehicles (IoV)
Full text linkThe Internet of Vehicles (IoV) integrates with different nodes, like for example connected vehicles, roadside units, etc. Due to communication exchange, they are exposed to various attacks on the network, which poses a security risk. Nevertheless, security is a major concern in IoV networks, especially during data transmission. To address this issue, our team suggest an innovative approach. reputation management schema in an IoV environment to detect attacks at an early stage based on vehicle and driver behavior along with network state. Our algorithm combines direct and indirect trust with various metrics like Packet Lost Rate (PLR), vehicle speed distance between neighbors, alert content, and link quality. These metrics are used to compute a reputation score to identify malicious nodes. Based on its reputation, vehicles communicate with only trusted nodes. After assessment, we see that our solution surpassed the others solution and has demonstrated superior effectiveness in detecting abnormal vehicles. Furthermore, the computed delay, equal to 4.7 ms, does not affect the network communications, which is interesting for the introduced safety features
Performance of Machine Learning Algorithms on Automatic Summarization of Indonesian Language Texts
Full text linkAutomatic text summarization (ATS) has become an essential task for processing huge amounts of information efficiently. ATS has been extensively studied in resource-rich languages like English, but research on summarization for under-resourced languages, such as Bahasa Indonesia, is still limited. Indonesian presents unique linguistic challenges, including its agglutinative structure, borrowed vocabulary, and limited availability of high-quality training data. This study conducts a comparative evaluation of extractive, abstractive, and hybrid models for Indonesian text summarization, utilizing the IndoSum dataset which contains 20,000 text-summary pairs. We tested several models including LSA (Latent Semantic Analysis), LexRank, T5, and BART, to assess their effectiveness in generating summaries. The results show that the LexRank+BERT hybrid model outperforms traditional extractive methods, achieving better ROUGE precision, recall, and F-measure scores. Among the abstractive methods, the T5-Large model demonstrated the best performance, producing more coherent and semantically rich summaries compared to other models. These findings suggest that hybrid and abstractive approaches are better suited for Indonesian text summarization, especially when leveraging large-scale pre-trained language models
A Data Science Approach to Exploring the Relationship Between TikTok Engagement and Revenue in Malaysia: A Case Study of the Beauty and Personal Care Sector
Full text linkTikTok has reshaped digital marketing in the beauty and personal care sector, yet the relationship between engagement metrics and revenue outcomes remains unclear. This study aims to examine how public engagement metrics (likes, comments, shares, and live interactions) relate to revenue performance among TikTok influencers. Using the Data Science Trajectories (DST) framework, data from 17 Malaysian influencers across Celebrity, Macro, Meso, and Micro categories were analyzed through descriptive statistics and machine learning models implemented in Python. The findings reveal that high engagement does not consistently lead to higher revenue. Live sessions were more effective than standard videos in driving sales due to real-time interaction. While Celebrity influencers led in revenue, Meso influencers recorded the highest engagement rates. A Random Forest regression model showed strong predictive power (R² = 0.94), demonstrating that public-facing metrics can be used to estimate revenue. The study also introduces category-based engagement rate benchmarks and highlights the unique value of live content in converting engagement into sales. This research contributes to the growing body of work on TikTok marketing by combining statistical and predictive techniques to link engagement behavior with commercial outcomes, offering actionable insights for both practitioners and scholars
Blockchain-Enabled Secure Healthcare Data Management with Modified Gazelle Optimization and DLT-Trained RNN-BILSTM Approach
Full text linkThe growth of the healthcare system has posed challenges in safeguarding patient privacy amidst the storage, distribution and management of medical data. Blockchain (BC) offers a promising result by securely enabling the exchange of medical information. Utilizing block chain technology ensures the security of individuals\u27 confidential health information. The use of a decentralized, immutable ledger using blockchain technology provides a secure, impenetrable platform for storing and retrieving private medical information, protecting patient privacy. The application of Modified Gazelle Optimization enables the determination of the shortest path for efficient data transfers within the block chain network. By adopting a specialized routing protocol called Modified Gazelle Optimized Routing, this approach minimizes latency and maximizes throughput, facilitating continuous and expedited transfer of health data across the network. To assure the data confidentiality and integrity of network nodes, a Distributed Ledger Technology (DLT) trained Recurrent Neural Network with Bidirectional Long Short Term Memory (RNN-BILSTM) approach is implemented. This advanced Deep Learning (DL) technique enhances the security and reliability of the network by detecting and preventing unauthorized access and tampering attempts. The proposed RNN-BILSTM based Intrusion Detection System (IDS) efficiently detects different types of attacks with high accuracy. By analyzing network traffic and patterns in real-time, the IDS have the ability to identify and mitigate harmful Internet of Things (IoT) requests and various stealthy attack types, including previously unknown threats. The outcomes of this research prove an efficacy and consistency of the proposed strategy in enhancing the security, efficiency and performance matrix with an accuracy of 97% and comparative analysis is done with traditional methods, thereby ensuring an availability and integrity of healthcare data