Performance Comparison of PSO, ABC, and ACO Algorithms For Multi-UAV 3D Path Planning and Collision Avoidance Reza Manazil, Nyoman Karna, Soo Young Shin Proceedings of Icitda 2025 10th International Conference on Information Technology and Digital Application, 2025 Multi-UAV (Unmanned Aerial Vehicle) systems have shown great potential in various fields, yet their operation faces critical challenges in path planning and collision avoidance. Swarm Intelligence (SI) algorithms such as PSO, ABC, and ACO offer promising solutions, but a comprehensive performance comparison of these three in 3D scenarios remains limited. This research aims to systematically analyze and compare the performance of the Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC), and Ant Colony Optimization (ACO) algorithms in finding safe and efficient path solutions. The comparison was conducted through computational simulations in two different density scenarios, namely 10 UAVs and 30 UAVs, by evaluating metrics of computation time, number of turns, and the closest distance between UAVs. The results indicate that no single algorithm is absolutely superior. PSO consistently proved to be the fastest in terms of computation time, with an average of 61.01 seconds in the 10-UAV scenario. However, ABC demonstrated advantages in path quality by producing the fewest turns (average of 26.5) and was the most reliable in maintaining safe interUAV distances, especially in the high-density 30 UAV scenario where it maintained an average distance of 4.715 meters. Based on these findings, it can be concluded that the selection of the best algorithm is highly dependent on mission priorities: PSO is suitable for missions requiring rapid solutions, while ABC offers superior reliability and safety for more complex operations.
Lightweight Deep Learning Model for Drowsiness Detection Ridho Al Harits, Nyoman Karna, Inung Wijayanto Icons Iot 2025 International Conference on Networking Intelligent Systems and Iot, 2025 Traffic accidents remain a significant public health concern, causing substantial material losses and fatalities. To mitigate this issue, researchers have developed systems to monitor driver alertness and provide early warnings when signs of drowsiness are detected. However, most state-of-the-art drowsiness detection models rely on high-performance computing devices, which limits their applicability in low-cost, energy-constrained environments. This study investigates a lightweight deep learning solution using MobileNetV2 for eye-state-based drowsiness detection. The model is trained on a dataset of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$96 \times 96$</tex> grayscale images using a pre-trained MobileNetV2 architecture and converted into TensorFlow Lite (TFLite) format for deployment on edge devices. To further optimize the model for embedded use, quantization into Float32 and Float16 formats was performed. Experimental results show that the proposed model achieves high accuracy with only 93,234 parameters, offering a 97% reduction in parameter complexity compared to prior work, with only a 1% decrease in accuracy. These findings highlight the potential of the proposed approach for real-time drowsiness detection on lowpower devices.
Intrusion Prevention System Optimization Using Machine Learning Ikram Andika Ukar, Nyoman Bogi Aditya Karna, Danu Dwi Sanjoyo 2025 International Conference on Converging Technology in Electrical and Information Engineering Iccteie 2025, 2025 The growing reliance on networked systems continues to increase the risk of cyberattacks, particularly denial-of-service (DoS) and distributed denial-of-service (DDoS) threats targeting IoT-enabled infrastructures. Traditional Intrusion Prevention Systems (IPS) such as Snort depend on static signatures, making them limited in detecting novel or application-layer attacks. This study evaluates a machine learning-based IPS deployed on a Linux cloud server and trained on the CIC-IoT2023 dataset. Among multiple classifiers, the Decision Tree (DT) algorithm was selected for its balance of accuracy, efficiency, and interpretability. DT achieved the highest performance with 99% accuracy and F1-score on SYN Flood, Slowloris, and Benign traffic classes. More importantly, in live attack simulations using Docker-based attacker nodes routed through VPNs, the ML-IPS successfully blocked all malicious attacker IPs in near real time, as confirmed by iptables logs. The system maintained low resource usage (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$<10 \%$</tex> CPU) and sub-second detection latency, demonstrating both scalability and robustness. These findings highlight the suitability of Decision Tree models for enhancing IPS performance in IoT-oriented environments, enabling accurate detection while ensuring effective prevention.
Evaluating Azure Site Recovery for Disaster Recovery Yulia Vironica, Sofia Hertiana, Nyoman Karna Proceedings International Seminar on Intelligent Technology and Its Applications Isitia, 2025 Azure Site Recovery (ASR) is a cloud-based disaster recovery solution that enables organizations to maintain business continuity by automating data replication, failover, and failback processes. This paper comprehensively evaluates the effectiveness of ASR in achieving critical disaster recovery objectives, particularly Recovery Time Objectives (RTO) and Recovery Point Objectives (RPO), through real-world simulations involving Linux-based virtual machines deployed in Azure environments. The experimental setup included multiple failover and failback scenarios with varying workloads, monitored using Azure Monitor and Log Analytics. Findings reveal that ASR consistently delivers RTOs below 30 seconds and RPOs within five minutes, indicating strong performance in minimizing downtime and data loss. Key operations such as Finalize Protection and Unplanned Failover showed high reliability, with success rates above 95 %. The study also discusses potential optimization areas in large-scale data replication and Azure-native failover strategies. This evaluation provides practical insights and evidence-based recommendations for organizations seeking scalable and efficient cloud-based disaster recovery solutions.
Generalization Performance of Internet of Things Intrusion Detection System Built on Impact-based Dataset Using TabNet Architecture Aldira Fadillah Lazuardi, Suryo Adhi Wibowo, Nyoman Karna 2025 International Conference on Data Science and Its Applications Icodsa 2025, 2025 Internet of Things (IoT) networks consist of numerous small devices that are interconnected, gathering and transmitting data from one another to generate information on specific subject. Advancements of IoT have reached almost all aspects of modern life, ranging from simple room monitoring devices to industrial applications designed to streamline production and boost productivity. One subsequent factor of this rapid advancement is vulnerable devices placed on networks, creating a hole for malicious intrusions. One method in battling intrusions is Intrusion Detection Systems (IDS). This study presents a generalization analysis of a model created with TabNet trained on the CIC IoT-DIAD 2024 dataset. And for generalization, the model is tested against CIC IDS 2017, a popular dataset for IDS classification. The model was able to achieve decent results, achieving 84.39% and 91.80% F1-score for multiclass and binary classification, respectively. In contrast, only 50.29% and 21.26% F1-score was achieved for generalization for multiclass and binary classification, thus making the results still poor. The results for classification were promising, this can help further improve research on TabNet as an architecture for IDS.
Internet of Things Network Security with Intrusion Prevention System Based on SnortML Machine Learning Fajar Hadi Hidayatullah, Nyoman Karna, Favian Dewanta Proceedings 7th International Conference on Informatics Multimedia Cyber and Information System Icimcis 2025, 2025 The potential risks associated with cyber-threats posed by IoT devices are significant and diverse, e.g., endangering users, causing asset losses, to damaging the physical ecosystem. This research aims to develop an appropriate Intrusion Prevention System (IPS) to protect IoT networks in MSMEs, which have limited budgets for network security, and invest in their employees’ security knowledge in security is nearly impossible. At the same time, the risk imposed by cyber-attacks is as high as larger companies. By integrating Raspberry Pi with SnortML in Snort, this is expected to be an alternative solution for MSMEs to detect and prevent cybercrime, especially DDoS and exploit attacks from reconnaissance to exfiltration. The research method includes implementing Snort software and employing Convolutional Neural Networks (CNN) and Long Short Term Memory (LSTM) to enhance detection accuracy on Raspberry Pi, such as Raspberry Pi 3B and Raspberry Pi 4; on both devices, we then analyze resource utilization during the attack experiments and evaluate the model’s metrics. Evaluations are conducted using confusion matrix metrics (accuracy, precision, and recall), along with resource utilization assessments (CPU and RAM). The model LSTM and CNN on Raspberry Pi 3B have an accuracy of 60.39% and 59.08%, while in Raspberry Pi 4, the accuracy is 61.61% and 65.86%. This paper chose CNN with Raspberry Pi 4 as the best model because it resulted in an accuracy of 65.86%.
Uplink RSMA-Assisted Slotted ALOHA With Adaptive Traffic Load for Massive IoT I Nyoman Apraz Ramatryana, Gandeva Bayu Satrya, Nyoman Bogi Aditya Karna, Made Adi Paramartha Putra IEEE Communications Letters, 2025 This letter presents a design of preconfigured signal-to-interference-plus-noise ratio (PSINR) level allocation in an uplink rate splitting multiple access (RSMA)-assisted slotted ALOHA (RSMA-ALOHA) for massive internet of things (IoT). However, considering massive IoT under high traffic load, the throughput of RSMA-ALOHA is degraded due to collisions. To solve this, adaptive traffic load (ATL) is proposed to manage high traffic load and stabilize throughput. The ATL mechanism dynamically controls the connectivity of massive IoT devices according to the traffic load estimation. The throughput bounds are derived, demonstrating that RSMA-ALOHA with ATL significantly outperforms the benchmark. Simulation results validate the theoretical analysis, showing that RSMA-ALOHA maintains superior throughput performance under high traffic loads with the cost of delay.
Classification of SO2 Emissions in Indonesia Paiton Coal-Fired Power Plant Area Based on Screening Dust AOT Data Susan Agustia, Jangkung Raharjo, Inung Wijayanto, Nyoman Bogi Aditya Karna Icons Iot 2025 International Conference on Networking Intelligent Systems and Iot, 2025 The classification of aerosol spikes in the environment around coal-fired power plants is essential for understanding environmental pollution and the urgency of early detection and mitigation activities. Existing aerosol classification methods cannot balance model performance and computational efficiency, nor can they separate emissions from coal plants or dust emissions. In this paper, we propose a new approach to model the filtering of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{SO}_{2}$</tex> emissions that occur either purely from coal-fired power plant emissions or from exposure to other emissions. Using an artificial neural network based on satellite image data, the proposed classification is able to filter out the source of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{S O}_{2}$</tex> emissions occurring in the area of Paiton Coal-fired Power Plant, the largest coal-fired power plant in Indonesia. The screening interaction mechanism is performed by utilizing Dust Aerosol Optical Thickness (Dust AOT) data from Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2) Analysis as the emission source determinant. This approach increases learning efficiency and improves classification performance with a correct assessment of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{SO}_{2}$</tex> emissions in the research environment. The proposed model achieved a classification accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 9. 1 \%}$</tex> in the Random Forest algorithm with 3 datasets, namely the Aerosol Optical Thickness (AOT) index of Particulate Matter <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2.5\left(\text{PM}_{2.5}\right)$</tex>, the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{SO}_{2}$</tex> emission index, and the Dust Aerosol Optical Thickness (Dust AOT) index. The model outperforms other models because it can effectively capture the dependency on mixing emissions with dust pollution that does not originate from coal-fired power plants. This highlights the advantages of the proposed method in aerosol emission classification.
Investigating the Effect of Different Augmentation Strategies on MobileNetV2 for Strawberry Disease Classification Sazqia Aulia Palad, Haikal Febriano, Ledya Novamizanti, Nyoman Bogi Aditya Karna Proceedings of the 2025 IEEE International Conference on Industry 4 0 Artificial Intelligence and Communications Technology Iaict 2025, 2025 Strawberries are a key agricultural product, appreciated for their nutritional richness, particularly in potassium, vitamin C, phosphorus, calcium, and magnesium. Nonetheless, strawberry farming often encounters difficulties due to diseases, pests, and weeds, which can significantly reduce yield and lower fruit quality. This study explores how different data augmentation methods impact the classification performance of MobileNetV2, a lightweight convolutional neural network, in detecting strawberry diseases. A dataset containing 1,200 images across six disease categories was used, applying three augmentation levels: minimal, moderate, and aggressive. Results revealed that moderate augmentation achieved the best outcomes, with an accuracy of $98 \%$, a loss of 0.05, and perfect scores for precision, recall, and $F 1$. In contrast, the aggressive augmentation yielded a slightly reduced accuracy of $97 \%$ and a higher loss of 0.27, possibly due to excessive distortion. Meanwhile, the minimal augmentation strategy also performed well, achieving $96 \%$ accuracy with a 0.09 loss. These results highlight the critical role of selecting suitable augmentation strategies to enhance generalization and ensure robust performance in practical strawberry disease identification scenarios.
Loss-Based Decentralized Federated Learning for Robust IoT Intrusion Detection System Made Adi Paramartha Putra, Nengah Widya Utami, I Gede Juliana Eka Putra, Nyoman Karna, Tia Rahmawati, Rama Wijaya Shiddiq, Ahmad Zainudin, Gabriel Avelino R Sampedro Proceedings of the 2024 IEEE International Conference on Industry 4 0 Artificial Intelligence and Communications Technology Iaict 2024, 2024
Efficiency and Effectiveness of Water Sprinkler Usage in Balinese Agriculture Dewa Rahyuni, Nyoman Karna, Sofia Hertiana, Sussi, I Nyoman Ganeshan Ananda Putra, I Wayan Risko Surya Cita, I Kadek Andika Herlantika, Made Adi Paramartha Putra 2024 Asu International Conference in Emerging Technologies for Sustainability and Intelligent Systems Icetsis 2024, 2024
Prototype of Chilli Plants Automation System in IoT-Based Smart Greenbox Ikram Andika Ukar, Nyoman Karna, I Putu Yowan Nugraha Suparta Icacnis 2022 2022 International Conference on Advanced Creative Networks and Intelligent Systems Blockchain Technology Intelligent Systems and the Applications for Human Life Proceeding, 2022
Decision Tree-Based Bok Choy Growth Prediction Model for Smart Farm Aldi Sulthony Susilo, Nyoman Karna, Ratna Mayasari Icoiact 2021 4th International Conference on Information and Communications Technology the Role of AI in Health and Social Revolution in Turbulence Era, 2021
Executive dashboard as a tool for knowledge discovery Nyoman Karna Proceedings 2017 International Conference on Soft Computing Intelligent System and Information Technology Building Intelligence Through Iot and Big Data Icsiit 2017, 2017
Knowledge sharing filtering on OAI-PMH Rika Yuliant, Nyoman Karna 2016 International Conference on Information Technology Systems and Innovation Icitsi 2016 Proceedings, 2017
Social CRM using web mining Nyoman Karna, Iping Supriana, Ulfa Maulidevi 2014 International Conference on Information Technology Systems and Innovation Icitsi 2014 Proceedings, 2014
