Evaluation of AEAD Cryptographic Algorithms in an MQTT-Based IoT Tracking System Geva Almer Hariri, Favian Dewanta, Sri Astuti Proceedings 2025 8th International Seminar on Research of Information Technology and Intelligent Systems Isriti 2025, 2025 Internet of Things (IoT) technology has become widely utilized in various domains, including real-time location tracking systems that rely on sensor data and wireless communication. However, transmitting sensitive information over public networks raises serious security concerns. To address these issues, AEAD (Authenticated Encryption with Associated Data) algorithms such as AES-GCM, ChaCha20-Poly1305, and Ascon are implemented to ensure both data confidentiality and integrity. This study evaluates five main aspects which include encryption and decryption processing time, the efficiency of the encrypted payload size, the effectiveness of data obfuscation based on Wireshark packet capture analysis, avalanche effect strength, and memory utilization. The results show that ChaCha20-Poly1305 delivers the fastest performance, Ascon produces the smallest payload and requires the least memory, while AES-GCM offers the strongest avalanche effect with a balanced trade-off between speed and data size. All algorithms successfully obscure message contents, making them suitable for secure and efficient IoT-based tracking systems.
Physical Activity Classification on DST Devices Based on Accelerometer Data Using Machine Learning in an Encrypted IoT System Mohamad Farrel William Rosyadi, Favian Dewanta, Sri Astuti Proceeding of the International Conference on Computer Engineering Network and Intelligent Multimedia 2025 Cenim 2025, 2025 Children with Down Syndrome have communication limitations and difficulties in maintaining spatial orientation during outdoor activities, requiring a system that can track position and detect physical activities in real time. Most existing tracking devices lack automatic classification of activities from motion data. This study classifies physical activities based on accelerometer data from an MPU6050 sensor on an ESP32-based IoT device, transmitted in real time via MQTT. A moving average method with window sizes of 5,10, and 15 was applied to improve data stability. The dataset contains three features ($\mathrm{X}, \mathrm{Y}, \mathrm{Z}$) and four classes: stationary, walking, running, and in a vehicle. In the testing phase, all models maintained accuracy levels around $80-83 \%$ under different data splits (90:10, 80:20, and 70:30), indicating good generalization performance.
Comparative Analysis of Neural Networks and Traditional Machine Learning Models in Federated Learning for Multi-Class Classification Sri Astuti, Aloysius Adya Pramudita, Favian Dewanta Proceedings 2025 8th International Seminar on Research of Information Technology and Intelligent Systems Isriti 2025, 2025 Machine learning continues to develop rapidly. The utilization of data training with centralized machine learning models in multi-class classification tasks has been demonstrated to yield significant effectiveness. Considering prevailing privacy concerns, federated learning has emerged as a salient solution. Federated Learning (FL) is a new paradigm in machine learning that enables distributed model training across clients without sharing raw data and while maintaining data privacy. However, its effectiveness in multi-class classification tasks faces several challenges compared to existing centralized machine learning approaches, such as data heterogeneity and computational constraints. This study compares the performance, computational efficiency, and stability of Neural Networks, SGD Classifier, Logistic Regression, and Random Forest within the FL framework for multiclass classification on a 16-class dataset, using the Flower framework. We evaluate these models in 10 rounds with three clients. This study focuses on accuracy, loss, training time, and robustness to non-IID data distributions. The findings of this study demonstrate that artificial neural networks achieve the highest level of accuracy (98.67%) with a loss of 0.0440, thus demonstrating optimal performance and stability. Conversely, Random Forest demonstrates a high level of accuracy (97.34%) and exceptional efficiency (an average runtime of 1.61 seconds). However, it is observed to exhibit instability. Logistic regression (91.92% accuracy) and SGD classifier (92.69% accuracy) are computationally efficient, but their effectiveness is limited due to their linear nature and instability. The findings of this study offer practical insights regarding model selection in the domain of FL applications. Furthermore, they emphasise the necessity of optimised strategies for the purpose of addressing data heterogeneity and computational trade-offs.
Integration of Internet of Things, Machine Learning, and Mobile Applications for Location Tracking and Activity Prediction of Individuals with Down Syndrome Moch Firza Yudistira Meizia, Favian Dewanta, Sri Astuti Proceeding of the International Conference on Computer Engineering Network and Intelligent Multimedia 2025 Cenim 2025, 2025 Individuals with Down Syndrome often experience disorientation and limited communication abilities, making it difficult for them to navigate safely in daily life. To address these challenges, an IoT-based tracking system was developed to enhance their safety and independence. The system integrates a wearable IoT device, a cloud-hosted MQTT broker, and a Flutter mobile application to transmit and display GPS, accelerometer, temperature, and humidity data in real time. A pre-trained Support Vector Machine (SVM) model classifies activities walking, running, or traveling in a vehicle while AES-GCM encryption ensures data confidentiality and integrity. Testing recorded an average transmission latency of 2.096 ms, SVM accuracy of 90% at a window size of 5, and average encryption/decryption times of 0.3 ms and 0.39 ms, demonstrating reliability and security for continuous real-time monitoring.
Sec-DST: Secure IoT-Based Tracking System for Children with Down Syndrome Favian Dewanta, Retno Hendryanti, Ahmad Tri Hanuranto, Sofia Naning Hertiana, Sri Astuti Proceeding of the IEEE International Conference on Smart Instrumentation Measurement and Applications Icsima, 2025 Children with down syndrome require special care due to their unique physical and mental development, especially when interacting with people and the environment around them. Although Internet of Things (IoT)-based tracking systems offer promising solutions to ensure their safety and well-being, these systems can introduce significant security and privacy concerns that may expose children to potential threats from malicious actors. This paper presents a secure IoT-based tracking system for children with down syndrome (Sec-DST). This scheme is designed to protect the communication channels between IoT tracking devices and the mobile phone while preserving the privacy of children with down syndrome against attackers and criminals who may attempt to track their movements. Security analyses show that the Sec-DST approach can protect message queueing telemetry transport (MQTT) messages between IoT devices and mobile applications against well-known attacks. In addition, the experimental results demonstrate that the computational cost of Sec-DST is around 40 ms for both the SHA1 and SHA256 hash functions with a plaintext size of 100-bytes in the ESP32 hardware platform.
Abstractive Indonesian Text Summarization Model Using BERT and GPT-2 Architecture Muhammad Karov Ardava Barus, Bagas Eko Tjahyono Putro, Intan Nisa Bani, Alfarelzi, Sri Astuti, Ibnu Asror Proceedings of 2025 IEEE International Symposium on Future Telecommunication Technologies Softt 2025, 2025 This paper presents a comprehensive study on the effectiveness of the BERT2GPT model in addressing the growing demand for efficient summarization of large-scale textual data, with a particular focus on abstractive summarization methods. Summarization plays a critical role in enabling users to quickly digest important information from lengthy documents, especially in the era of big data where the volume of digital text continues to expand rapidly. Unlike extractive techniques, which rely on directly selecting key sentences from the source text, abstractive summarization involves generating entirely new sentences while still preserving the core meaning and essential context of the original content. This makes abstractive approaches more challenging, but also more powerful in producing natural and human-like summaries.The BERT2GPT model combines the strengths of two well-established architectures: BERT and GPT-2. BERT contributes its bidirectional contextual understanding, allowing the model to grasp nuanced meanings from the source text, while GPT-2 provides robust generative capabilities, enabling the production of coherent and contextually appropriate summaries. To ensure relevance in the Indonesian language domain, the model was fine-tuned on local datasets, specifically IndoSum and Liputan6, which consist of news articles widely used in summarization research.Evaluation of the fine-tuned model demonstrates promising results. The BERT2GPT framework achieved average ROUGE-1, ROUGE-2, and ROUGE-L scores of 0.67, 0.58, and 0.64 respectively on the IndoSum dataset. These scores indicate significant improvements in both coherence and informativeness of the generated summaries compared to baseline methods. Overall, the findings suggest that BERT2GPT can be effectively utilized for summarizing large volumes of Indonesian digital content, providing valuable support for applications such as news aggregation, digital libraries, and educational platforms.
Text Classification Using NLP by Comparing LSTM and Machine Learning Method Zaidan Muhammad Mahdi, Retno Fauziah Istiqomah, Alfarelzi, Sri Astuti, Ibnu Asror, Ratna Mayasari Proceeding of 2024 the 10th International Conference on Wireless and Telematics Icwt 2024, 2024 Natural Language Processing (NLP) has seen significant advancements recently, leading to various applications across different domains. This research focuses on text classification by comparing the performance of Long Short-Term Memory (LSTM) networks with several traditional machine learning algorithms, including Naive Bayes, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and XGBoost. The novelty of this study lies in its comprehensive comparative analysis using a balanced dataset. The data balancing was achieved through the Synthetic Minority Over-sampling Technique (SMOTE), ensuring robust model training. Experimental results reveal that the SVM algorithm achieves the highest accuracy of over 96%, surpassing other models in performance. This indicates that despite the advancements in deep learning, traditional algorithms like SVM remain highly effective for text classification tasks. The findings provide valuable insights into the strengths and weaknesses of different NLP approaches, contributing to the ongoing development of more accurate and efficient text classification models. Future work could explore hybrid models and it application to diverse datasets to further enhance classification accuracy.
Integration of Software-Defined Networking with Named Data Network for Implementing Forwarding Strategies in Wireless Networks Reza Maharani Susilo, Farraz Rizky Kusumaputra, Muhammad Hendrawan Adiwijaya, Ratna Mayasari, Ridha Muldina Negara, Sri Astuti 6th International Seminar on Research of Information Technology and Intelligent Systems Isriti 2023 Proceeding, 2023 Named Data Networking (NDN) represents a forward-looking networking concept that addresses various challenges within the current internet architecture, particularly the reliance on IP addresses for data transmission between devices. In response, Named Data Networking (NDN) and Software-Defined Networking (SDN) architectures introduce a novel approach to data delivery by shifting from a host-centric to a data-centric model. This transition not only enhances data distribution efficiency but also leverages SDN advantages stemming from its segregation of the data and control planes. To implement this convergence, we incorporated the SDN paradigm into the NDN environment. By doing so, we harnessed the capabilities of both SDN and NDN, enhancing network efficiency and reducing data retrieval time for consumers. We utilized Named Data Link State Routing (NLSR) as a routing protocol within the default NDN environment. However, NDN encompasses diverse forwarding strategies tailored to specific network conditions. This research specifically investigates Best-Route Forwarding, Multicast Forwarding, and Adaptive Smoothed RTT Forwarding strategies. The objective is to evaluate the disparities and appropriateness of these strategies within NLSR-NDN and SDN-NDN environments when applied to wireless networks. To assess the efficacy of these strategies, our analysis employs Quality of Service (QoS) parameters, encompassing Average Round Trip Time (RTT), Throughput, Packet Loss, and Satisfied Interest Ratio. These additional metrics provide a comprehensive evaluation of interest satisfaction. Our findings reveal that the SDN-NDN environment remarkably enhances network efficiency by approximately 50-70% compared to the NLSR-NDN environment.
Forwarding Strategy Analysis in Wireless Network Based Named Data Network (NDN) Reza Maharani Susilo, Farraz Rizky Kusumaputra, Muhammad Hendrawan Adiwijaya, Ratna Mayasari, Ridha Muldina Negara, Sri Astuti Proceedings Ieit 2023 2023 International Conference on Electrical and Information Technology, 2023 Along with the time, the development of the internet has grown so rapidly that it cannot be controlled. This has resulted in the current internet architecture that no longer being able to meet existing needs. The emergence of Named Data Networking (NDN) architecture can help to overcome previous problems. However, NDN architecture also has its own shortcomings or problems such as Broadcast Storm, which causes packets to be sent in the network to spin continuously that make requiring more energy and consuming considerable time. In this final project, the author implements a comparison of Forwarding Strategy methods focused on Analysis of Forwarding Strategies in Wireless Network Based Named Data Network to find the Best Strategies to prevent the Broadcast Storm and improve the efficiency of Network. Performance testing is carried out by experimenting with several different Forwarding Strategies. After experimenting with the Forwarding Strategies, the authors can find which is the best Forwarding Strategy to prevent the Broadcast Storm that occurs in Wireless Networks.
Optimizing Forwarding Strategies in Named Data Networking Using Reinforcement Learning Zhafirah Naghmah Ahmad, Fika Triana, Revita Rachel, Ridha Muldina Negara, Ratna Mayasari, Sri Astuti, Syamsul Rizal Proceeding of 2023 9th International Conference on Wireless and Telematics Icwt 2023, 2023 In the current network architecture, IP addresses are used, where data transmission uses the host address on each device. From this data delivery method, NDN emerges as a new paradigm in data transmission from being host-centric to becoming data-centric. There is a strategy used in research with the weakness of congestion in the forwarding strategy. Therefore, modeling the forwarding strategy using Reinforcement Learning is designed to overcome this problem. In the run simulation, an environment will be created in the Reinforcement Learning system with several scenarios in the NDN network. To measure the success of the system, testing is carried out to achieve maximum results, such as the Reinforcement Learning process, which is trial and error in nature, which means that several experiments are carried out, such as the exploration process carried out by the agent in the environment to achieve the goal and get the expected maximum reward. The components used in Reinforcement Learning in the training process are agents, actions, policies, and rewards. The tests aim to make NDN an efficient network system, simplify network performance automatically using Reinforcement Learning, and make NDN a network system that can overcome congestion for forwarding.
