Indonesia ranks third globally in e-Health utilization, which is carried out by 57% of respondents. It shows that public awareness of the importance of technology in the health sector has changed significantly. This awareness needs to be continuously enhanced. However, this is not in line with the use of health services for pregnant women. This is related to the factor of adopting health technology that is in line with the features of e-Health itself. This research examines sentiment analysis on 17 e-Health in Indonesia to gain in-depth insight into user perceptions and responses related to features, especially Technical Factors and Security/Privacy Factors taken through the Google Playstore. The methods include sentiment labelling with TextBlob and modelling using Convolutional Neural Network (CNN). The overall model produced an accuracy of 71.67%, indicating that the model is most reliable in classifying aspects of Perceived Ease of Use and Performance Risk. This research is expected to provide valuable insights for application developers and the government in optimizing the quality and trust in e-Health in Indonesia.
Precise and prompt product identification is vital for efficient inventory management in the retail grocery industry. While deep learning models have significantly improved in classification accuracy, dynamically changing environments, seasonal variations, and new products pose significant challenges. These factors create new data on which pre-trained models are not adequately trained, leading to a substantial drop in accuracy. Continual learning has emerged as a solution, allowing models to adapt to new data without forgetting previously learned information. However, this solution still suffers from catastrophic forgetting, especially when dealing with large datasets. This paper aims to explore strategies to mitigate catastrophic forgetting and develop a robust model for fruit and vegetable classification in the retail sector. The models used are two memory-based models (DER++ and ER) and two regularization-based models (EWC and EWC + CPR). The result indicates that memory-based models outperform regularization-based models, with Dark Experience Replay++ (DER++) showing the highest accuracy in both classincremental and task-incremental settings. Memory-based models are less affected by catastrophic forgetting as they have direct access to past data, allowing them to have better knowledge preservation. Furthermore, the study also reveals that hybrid approaches (e.g., EWC + CPR) can further enhance retention accuracy.
School bullying remains an important issue affecting the safety and well-being of students. This study introduces a novel anti-bullying system that utilizes Internet of Things (IoT) sensors to predict and prevent bullying in educational Settings. The project consists of four main parts: introduction, literature review, proposed methodology, and project timeline with milestones. Through an extensive analysis of bullying in schools, including existing research and empirical data, the study identifies powerful strategies to mitigate such incidents. The proposed system combines voice recognition technology with the Internet of Things, utilizing hardware components such as Raspberry Pi, Arduino and various sensors to detect and respond to bullying. Voice sensors are strategically installed throughout the campus to identify and analyze sensitive speech and distress signals, immediately alerting school authorities for timely intervention. This integrated approach aims to enhance campus safety, ensure the physical safety of students and promote a healthy development environment. The success of the project relies on the joint efforts of schools, students, parents and the wider community to comprehensively address and reduce bullying. This paper Outlines the technical specifications, design, and implementation phases, and critically evaluates the effectiveness and limitations of the system.
Event Monitoring in Heterogeneous networks (HetNets) is extremely challenging due to the diverse network technologies and protocols involved. This study addresses the common challenge of event correlation in Heterogeneous Networks (HetNets). Which is crucial for ensuring network reliability and efficiency. In HetNets, we propose a novel way to integrate ServiceNow’s AIOps platform with advanced machine learning techniques to enhance correlation accuracy and reduce response times. Based on our findings, the proposed methodology outperforms existing solutions, resulting in significant improvements in both network performance and reliability. The research encompasses the integration of the ServiceNow platform with real-time HetNet of a big FinTech company. The platform ingests real-time data, advanced analytics, and predictive intelligence algorithms to detect, group, classify, and respond to network events. Key performance metrics, such as event detection accuracy, response time, and resource utilization, were systematically measured and analyzed. The findings indicate that the ServiceNow platform substantially enhances event correlation capabilities of HetNets. Significant improvements were observed in event detection accuracy and reduced response times when compared to distributed monitoring systems. The platform’s adaptability to different network configurations and its scalable architecture are important advantages. Additionally, the integration of machine learning techniques augments predictive intelligence capabilities, thereby reducing network downtime and enhancing automation and overall reliability. In conclusion, the ServiceNow platform’s advanced event monitoring mechanism provides a robust solution for heterogeneous networks. Future research will aim to further refine the machine learning models and expand the platform’s capabilities to support emerging network technologies.
The rail industry requires robust technology to support environmentally friendly and sustainable transportation, of which diesel generators are a key component. Currently, maintenance approaches for railway diesel generators tend to use preventive strategies that are less efficient and costly. Therefore, this study proposes the estimation of diesel generator health indicators to support predictive maintenance by utilizing multivariate stochastic approaches and exponential-based degradation models. This degradation model contains deterministic and stochastic parameters estimated using Bayes filter. In addition, to improve forecasting accuracy, this approach involves more than one type of measurement variable data and applies one of the multivariate analysis techniques, namely principal component analysis (PCA). The PCA variables are then used to determine the health indicators of diesel generators. As a case study, this research was conducted using data from a diesel generator on a power car. The results show that the health indicator is obtained when the tresshold has been reached.
Adaptive game design is a dynamic gamification approach that changes game elements such as challenges, feedback mechanisms, and rewards based on players’ preferences, behaviors, and needs. It is an emerging research field that aims to improve classic gamification techniques by adapting a game environment to meet the particular needs of various users and contexts. Machine learning (ML) techniques have the ability to predict future actions based on historical data to progress toward an ideal outcome. In digital gaming, adaptation of different gaming elements such as levels, difficulty, and feedback to match players’ skills is a fundamental requirement for an enjoyable experience. In this scenario, ML techniques can come into play. These techniques can help to advance the enjoyment of the gameplay toward players’ maximum satisfaction and engagement by anticipating the necessary changes in the game based on their previous behaviors. In this paper, we have conducted a systematic review of 17 papers to investigate the game balancing technology adopted recently. The fundamental goal of this survey is to provide a comprehensive overview of the latest developments in game balancing technologies that use ML techniques.
With the rapid proliferation of generative AI technologies, such as ChatGPT, and interoperable smart home standards like Matter, the smart home market is entering a new phase of development. The Matter standard, in particular, enables local control, allowing smart home services to function even without an internet connection. However, to leverage a wide range of AI services, including generative AI, user voice commands and data related to smart home devices must be transmitted to the cloud, which undermines the local control feature inherent to the Matter standard. To address this issue, this paper proposes a smart home system architecture that utilizes both a Matter Hub and a local AI server. We introduce the design of a Matter Hub that can control smart home devices and communicate with a local server, along with various APIs to facilitate data exchange between the hub and the local server. Experimental tests are conducted to evaluate the performance characteristics of the Matter Hub when used with a local server. Experimental results demonstrate up to a 320% improvement in device response times when using a local server. Additionally, extended performance tests reveal performance improvements of 321% on average, with up to 1492% in maximum response times and 320% in transactions per second (TPS) compared to cloud-based systems.
Current methods in mapping the availability of WiFi networks, such as crowdsourcing platforms (e.g. Project BASS and CoverageMap) and dedicated wardriving, face limitations in terms of data recency, volume, and cost-effectiveness. Due to the downsides of both these methods, opportunistic wardriving is proposed which utilizes public utility vehicles (PUVs). This study investigates the feasibility of PUVs as potential wardriving vehicles for opportunistic coverage mapping. This method is cost-efficient and removes the problem of data recency because of its continuous daily trips. The findings suggest that opportunistic wardriving with PUVs is viable for wardriving with its capability to detect a high volume of access points (APs). Multiple runs within the University of the Philippines Diliman campus showed the PUVs’ effectiveness in detecting a significant number of APs, with higher success rates at closer distances. Comparatively, warwalking detected fewer APs, but there was a significant overlap between the methods. The study also highlighted significant differences in detected AP types, with wardriving finding more mobile hotspots and miscellaneous devices than warwalking. Overall, the results underline the effectiveness of PUVs in providing extensive WiFi coverage data and further enhancements to the detection system could optimize the approach.
Social media became a primary means for individuals to share information and participate in social discourse in modern society. Analyzing tweets on Twitter provided a deep understanding of public opinion and reflected these insights in various policies. However, research on Japanese tweets related to the Ukraine-Russia war did not progress due to the lack of appropriate datasets and high API costs. In Japan, where pacifism is strong, exploring users' perceptions of this war provided crucial insights for future security policies. This study constructed a Japanese tweet dataset through meticulous preprocessing of a large multilingual tweet dataset from Kaggle. Multiple NLP techniques, including time-series analysis, co-occurrence analysis, LDA-based topic modeling, and BERT-based sentiment analysis, were applied to comprehensively analyze the war-related tweets. The analysis revealed that Japanese users showed a strong interest in the war, particularly concerning attacks on nuclear power plants and China's movements. Concerns about territorial issues and support for Ukraine indicated a recognition of national defense importance and potential involvement in an anti-Russia coalition. These insights provided critical perspectives for Japan's security policy planning. The need for stronger disinformation measures, attention to geopolitical risks, and emergency preparedness was highlighted. This study was one of the few data-driven studies that elucidated Japanese users' war perceptions. Future research should use broader data sources and advanced analytical methods to accurately capture national security trends from social media user interests.
In large-scale underwater wireless sensor networks, clustering of sensor nodes is often required to balance the lo ad and improve the throughput. However, due to the complexity of the hydroacoustic channel, existing clustering algorithms based on the distance between underwater nodes alone are not feasible. In this paper, we propose a multifactor clustering method MCHN based on channel quality in aquatic acoustic communication sensor networks, which simulates the aquatic acoustic channel quality through BELLHOP, and clusters nodes from four perspectives: inter-nodal communication quality, node com munication capability, network load balancing, and distance, which improves communication quality between nodes within the clusters and the clustering centre, and ensures load balancing to a certain degree in the network. that can be well applied to underwater wireless sensor network networks.
Forklifts are widely used in industrial and construction sectors for material handling, yet they pose significant safety risks due to blind spots, particularly at the rear of the vehicle. These blind spots are a leading cause of accidents involving forklifts, often resulting in severe injuries, especially among pedestrians. This study aims to enhance blind spot detection in forklifts by developing a system that integrates fuzzy logic with machine learning techniques. The proposed system incorporates critical variables such as steering angle and vehicle velocity to improve the accuracy of assessing severity of the obstacles within blind spots. Fuzzy logic is deployed to translate the expert or standard knowledge to machine. Three scenarios are analysed using Fuzzy logic Mamdani based on various sensors. Ultrasonic sensors were employed for close-range detection, while Artificial Neural Networks (ANNs) were used to analyze the frequency of dangerous states during forklift operations. The system's performance was evaluated across various scenarios, demonstrating its effectiveness in real-time safety assessments. The findings showst the potential to reduce the risk of accidents in forklift operations by providing timely alerts and accurate safety level predictions.
Data extrapolation via one-dimensional (1-D) parametric methods and a recurrent neural network (RNN) are investigated to study the evolution of the deforestation of Foret-des-Pins in Haiti. Satellite images collected by the National Aeronautics and Space Agency (NASA) from 1984 to 2022 are first used to extract entropy and correlation indexes and form time series from which losses in green space areas can be monitored. The Burg algorithm and state space are selected as parametric approaches, and the long-short term memory (LSTM) is chosen as the RNN to tackle the objective. Two extrapolation methods are considered: a traditional one, where the forecast is conducted via a one-step predicting scheme, and a second one with looping where each extrapolated sample is fed back to the input data, as time evolves, to compute the next one. Two experimental scenarios are conducted to judge the practicality of the extrapolation schemes, test the performance metrics of the three techniques, propose which one is best suited in each scenario and make a projection about the future of the forest if any mitigation measures are not adopted.
In recent years, natural language processing (NLP) has achieved notable advancements, particularly with large language models (LLMs) enhancing tasks such as sentiment analysis. Aspect-based sentiment analysis (ABSA), which involves identifying sentiment related to specific aspects within text, poses a more complex challenge compared to simple sentiment classification. To address this complexity, integrating multiple expert models has emerged as a promising approach. In this study, we propose the Evolutionary Expert Model merging with Task-adaptive Iterative Self-improvement Process (EEM-TISP) to improve ABSA task performance. EEM leverages evolutionary algorithms to merge expert models, optimizing task-specific outputs, while TISP iteratively refines these outputs, enhancing accuracy. Our results demonstrate that EEM significantly improves performance in simpler tasks, such as Aspect Term Sentiment Analysis (ATSA) and Aspect Category Sentiment Analysis (ACSA), surpassing individual expert models. TISP further boosts accuracy in these simpler tasks, though performance declines were observed in more complex ones. These findings indicate that while TISP is effective for simpler tasks, further optimization is required to tackle complex ABSA challenges, calling for future research into flexible prompt design and task-specific approaches.
In an age dominated by technology, manual tasks are becoming increasingly rare. However, gardening remains a beloved pastime for many, often requiring significant manual effort. The project aims to revolutionize gardening with the introduction of a smart plant watering system that minimizes manual labor and offers users unprecedented convenience. The system boasts a range of automated functionalities that can be scheduled for operation without human intervention or controlled remotely. Sensors, crucial in IoT systems, relay key information about environmental conditions necessary for plant health, such as light level, soil moisture, humidity, and temperature. The Particle Photon 2 microcontroller serves as the core processor, managing system operations and ensuring effective monitoring and control of the plant watering process. The result is an automated smart plant watering and monitoring system designed for small house plants, set to transform indoor gardening practices through IoT-enabled automation and monitoring. The goal of the project is to create a user-friendly system that seamlessly integrates automation, enhancing both efficiency and enjoyment for gardening enthusiasts.
This paper aims to model and implement a smart telemedicine service for students experiencing accidents or health issues on campus or in jogging areas. Unlike traditional systems based on uncertified wrist, finger, or arm measuring devices that merely inform the user and possibly their family and doctor, the proposed service utilizes a care model that combines both engineering and medical insights, adhering to telemedicine regulations set by the Ministry of Health. This approach aims to reduce costs for both medical staff and students and support AI-based alert and anomaly detection systems for timely care. The proposed service is a module of the S3 Campus project supported by EC Regional Funds aiming at developing a platform to provide security and video surveillance services, along with health support and professional visibility opportunities for students on a university campus.
In the context of increasing energy consumption and greenhouse gas emissions, this study explores the implementation of a smart building system using Internet of Things (IoT) technology at the Lontar Coal-Fired Power Plant (CFPP)’s administration building. The primary objective is to optimize energy usage without significant changes to existing infrastructure, aligning with ISO 50001 Energy Management System (EnMS) standards. An initial energy audit revealed high consumption, particularly from outdated air conditioning systems. The study highlights the innovative use of IoT devices, such as smart meters, sensors, and remote controls, enabling real-time monitoring and control of energy consumption. This approach provides a cost-effective solution for improving energy efficiency without requiring major infrastructure upgrades. Key findings include a 11.67% reduction in energy consumption, particularly in air conditioning and lighting systems, leading to substantial cost savings and reduced CO2 emissions. Furthermore, the project demonstrated the scalability of IoT for improving energy performance across large facilities. This study underscores the potential of IoT to drive energy efficiency, ensure compliance with environmental regulations, and offer a low-investment, high-impact solution for sustainable building management.
As the Internet of Medical Things (IoMT) continues to transform healthcare, it also introduces new vulnerabilities to sophisticated cyberattacks that outpace conventional defenses. In response, we present a tailored Intrusion Detection System (IDS) optimized for IoMT environments, designed to operate within the constraints of resource-limited devices while addressing complex, real-world attack vectors. Leveraging the CICIoMT2024 dataset and advanced machine learning models like Random Forest and XGBoost, our approach overcomes severe class imbalance and high dimensionality. Using Recursive Feature Elimination with Cross-Validation (RFECV), we reduced the feature set by 44.45%, achieving a state-of-the-art weighted F1-score of 99.48%. Despite the superior performance of the Random Forest model, its large memory footprint poses challenges for deployment on IoMT devices with limited resources. In contrast, the XGBoost model offers a better balance between high detection accuracy and resource consumption, making it more suitable for real-world applications. Our solution offers a scalable, efficient, and deployable IDS that brings a new level of adaptability and precision to IoMT cybersecurity, ready to defend against today’s threats while evolving to meet tomorrow’s challenges.
Indonesia, strategically positioned along the Pacific Ring of Fire, is exceptionally susceptible to seismic activities. Accurate estimation of ground motion parameters is crucial for mitigating the impact of earthquakes. Recent advancements in machine learning have expanded its application across various fields, including the analysis of seismic data. This research proposes the use of Artificial Neural Networks (ANNs) to enhance the accuracy and reliability of estimating ground motion parameters. Our methodology involves the collection and processing of data from seismic stations across Indonesia, provided by BMKG (Meteorological, Climatological, and Geophysical Agency of Indonesia). The dataset encompasses records from eight seismological stations, capturing three channels. This study primarily focuses on the Cianjur region, a zone highly prone to seismic disturbances, and intends to develop a robust model capable of accurately estimating ground motion parameters, particularly in regions with incomplete or limited seismic data. Therefore, this research is expected to contribute to ongoing efforts to improve the estimation of earthquake-related seismic parameters by estimating earthquake-related seismic parameters using the Artificial Neural Network (ANN) method and model optimization using the Particle Swarm Optimization (PSO) algorithm. The results were evaluated using PSO and without PSO. It is shown that the model using PSO shows better performance than without using PSO. In the ANN HNE + PSO Set 4 scenario, the model achieved an MSE of 0.185, MAE of 0.209, and a regression score of 0.904. The other scenarios generated in this study show good performance in general, but with variations in performance depending on the data set used.
Feature selection provides an accurate understanding of the problem under study. However, finding a stable and reproducible set of parameters that can win the confidence of experts in the field remains a challenge. In this study, we used a three-stage selection method to identify an efficient and robust group of variables that best describe the West African taurine cattle. Wrapper and embedded techniques were bagged and applied to 5 Machine Learning algorithms. The classification performance and robustness of the selection were assessed throughout the process. Finally, 12 out of 50 features were selected, and they were able to perform 96% precision as well as 97% accuracy, which is a significant improvement the results obtained on the initial dataset. The pairwise similarity and overall stability scores also increased. The resulting features were sufficiently robust to form a morphological prototype of lobi taurines.
Making decisions requires different information, especially when it is crucial. The traditional Analytic Hierarchy Process (AHP) is a great algorithm that will help make these decisions more quickly because it organizes the choices in a hierarchy. However, this algorithm has trouble with big datasets and changing decision settings because they depend on pairwise comparisons and subjective judgments when making decisions. Additionally, fuzzy and stochastic AHP algorithms were developed to improve the traditional ones. However, these algorithms are complex to use on a large scale and require many computers. This study presents a novel algorithm (Top-TC AHP) that integrates Support Vector Machine (SVM) and Top-K algorithms into the traditional AHP to help people make decisions faster and better. It uses SVM to sort data and get the probabilities of the criteria for the alternatives and Top-K to choose the top options. Time complexity, consistency ratio, and throughput were the metrics used to gauge the performance of the novel algorithm. As a result, when comparing Top-TC AHP to other ways, it gets better consistency, makes computing more manageable, and speeds things up. Because of this, it is beneficial for dealing with large amounts of data and situations that change quickly. In addition, the Top-TC AHP algorithm can help people make smart decisions in many areas, such as business and school planning.
