International Journal of Informatics and Communication Technology (IJ-ICT)
Not a member yet
494 research outputs found
Sort by
NLP-based fraudulent biomedical news identification using LSTM-SGD deep learning algorithm
Get PDFConcern over bio medical fake news is rising, particularly as false information about illnesses, medical procedures, and public health regulations becomes more prevalent. It is essential to recognize such false information, and deep learning (DL) algorithms can offer a potent remedy, especially when paired with sophisticated natural language processing (NLP) methods. This technique improves the model's capacity to ignore frequently used but uninformative terms and concentrate on important terminology. The model's capacity to concentrate on the most pertinent phrases for fake news identification is enhanced by the use of chi-squared, a statistical test that ascertains the dependency between various variables and aids in the removal of unnecessary data. By reducing less significant characteristics to zero, the Lasso approach, a kind of regression, is used for feature selection, guaranteeing that the model only utilizes the most predictive features for classification. A crucial step in getting the data ready for DL models is feature extraction, which turns unprocessed text into numerical data. After the structured data has been analyzed, algorithms like as stochastic gradient descent (SGD), long short-term memory (LSTM) may determine whether or not an article is accurate. The authenticity and dependability of medical information provided across platforms may be ensured by effectively identifying biomedical fake news by fusing DL with sophisticated NLP techniques
Plant disease sensing using image processing (with CNN)
Get PDFPlant disease is a significant challenge for agriculture, leading to reduced yield, economic loss, and environmental impact. Leveraging digital photos of plant leaves, convolutional neural networks (CNNs) have emerged as promising tools for disease detection. The methodology involves several steps, including image pre-processing, segmentation, feature extraction using CNNs. Crucially, a diverse dataset comprising images of both healthy and diseased leaves under varying conditions is necessary for training accurate models. Transfer learning, particularly with pre-trained models like ImageNet, can further enhance accuracy, allowing for better performance with fewer training samples. The proposed method demonstrates impressive results, achieving over 95% accuracy, outperforming existing state-of-the-art techniques. This system could serve as a valuable tool for farmers, facilitating timely disease identification and treatment, ultimately leading to increased agricultural yields, reduced financial losses, and the adoption of more sustainable farming practices. Additionally, beyond its practical applications, the proposed system holds promise for advancing sustainable agriculture by promoting environmentally friendly farming methods and contributing to the overall resilience and productivity of agricultural systems
Video-based physical violence detection model for efficient public space surveillance
Get PDFThis study aims to develop an effective real-time model for detecting violence in public spaces, focusing on achieving a balance between accuracy and computational efficiency. We evaluate various model architectures, with the main comparison between the ConvLSTM2D and Conv3D models commonly used in video analysis to capture spatial and temporal features. The ConvLSTM2D model, combined with preprocessing layers such as change detection and motion blur, showed optimal performance, achieving 86% accuracy after Bayesian optimization. With a low parameter count of 25,137, this model enables fast inference in just 0.010 seconds, making it suitable for real-time applications that require efficient computation. In contrast, the Conv3D model, which is also combined with preprocessing layers such as change detection and motion blur and has more than nine million parameters, shows a lower accuracy of 77.5% as well as a slower inference time of 0.025 seconds, making it unsuitable for real-time applications. The results of this study show that the ConvLSTM2D model is promising for real-time violence detection systems in public spaces, where a fast and accurate response is essential to prevent further acts of violence
Smart accommodation solution: innovative boarding house locator in Bayombong municipality
Get PDFThe search for affordable and conveniently located student accommodation is a common challenge, especially for students unfamiliar with their surroundings. This study presented the development and evaluation of a geographical information system (GIS)-enabled boarding house locator developed for Nueva Vizcaya State University (NVSU) students. The platform simplified the accommodation search process by providing a digital solution that integrates spatial data, real-time updates, and filtering options. The platform significantly reduced the time and cost of traditional housing searches. It helped students save 181.25 minutes per search and an average of 35 PHP in transportation costs compared to conventional methods like physical visits and word-of-mouth. Usability testing with 175 participants revealed high satisfaction, with the platform receiving an average rating of 4.83 for usability and 4.75 for performance. Key features such as interactive maps, location-based searches, and real-time updates enhanced the user experience by providing accurate, and up-to-date listings. The GIS-based platform outperformed traditional search methods in terms of efficiency and user satisfaction and offered a digital solution to common housing challenges faced by students. The results suggested the platform had strong potential for wider application at other universities. Overall, this system provides a scalable, cost-effective solution to improve student accommodation search and management
Privacy-preserving fitness recommendation system using modified seagull monarch butterfly optimized deep learning model
Get PDFThis paper presents a novel modified seagull monarch butterfly optimization (MSMBO) algorithm, with a multi-objective focus on privacy and personalization in the fitness recommender system using a refined three-tier deep learning structure. The method is divided into three phases. In the first phase, fitness data from wearable devices undergoes preprocessing to eliminate noise and standardize features. The second phase incorporates improved elliptic curve cryptography (IECC) alongside the MSMBO to encrypt user data securely, ensuring privacy in cloud storage. This phase also enhances neural network performance by optimizing weights and hyperparameters through feature selection, effectively reducing data complexity while boosting accuracy. In the third phase, ConvCaps extracts spatial data features, while Bi-LSTM identifies temporal dependencies. The proposed system balances multiple objectives like novelty, accuracy, and precision, while safeguarding user data through robust encryption. With the experimental findings, our suggested method performs better than current existing models, especially in heart rate prediction and fitness pattern identification. The overall outcome makes the system ideal for privacyconscious, personalized fitness recommendations. The model’s shows significant improvement in mean squared error (MSE), normalized mean squared error (NMSE), and mean absolute percentage error (MAPE), thus verifying its effectiveness in secure, real-time fitness tracking
GSM based load monitoring system with ADL classification and smart meter design
Get PDFThis paper introduces a method for the classification of activities of daily living (ADL) by utilizing smart meter and smart switch data in a synergistic approach. Through the integration of these internet of things (IoT) devices, the paper aims to enhance the application of ADL classification. Guided by recent advancements in load monitoring and energy management systems, the methodology incorporates machine learning techniques to analyze data streams from both the smart meter and smart switch. Drawing inspiration from prepaid smart meter monitoring systems, IoT-based smart energy meters for optimizing energy usage, and energy metering chips with adaptable computing engines, our design incorporates diverse perspectives. Additionally, we consider the utilization of mobile communication for prepaid meters, remote detection of malfunctioning smart meters, and an empirical investigation into the acceptance of IoT-based smart meters. We substantiate our proposed approach through experimental results, showcasing its effectiveness in accurately classifying diverse ADL scenarios. This research contributes to the field of smart home technology by offering an advanced method for ADL classification. The integration of smart meter and smart switch data provides a comprehensive understanding of energy consumption patterns, opening avenues for improved energy management and informed decision-making within smart homes
An integration clustering and multi-target classification approach to explore employability and career linearity
Get PDFThis study analyzes job placement waiting times and job linearity among female science, technology, engineering, and mathematics (STEM) graduates using clustering and multi-target classification (MTC) models. The K-means least trimmed square (LTS) algorithm, known for its robustness against outliers, was employed for clustering. With k = 2 and a trimming percentage of 30%, the model achieved a silhouette score of 77%, resulting in two distinct clusters: ideal and non-ideal. To enhance the dataset for classification, synthetic data was generated using the adaptive synthetic (ADASYN)-gaussian method. Principal component analysis (PCA) was used for visualization purposes, along with overlapping histograms, to illustrate that the synthetic data distribution closely resembled the original. For classification, a random forest (RF) model was used to predict both jobs waiting time and job linearity. Hyperparameter tuning produced an optimal model with a classification accuracy of 92%. Cross-validation (CV) confirmed the model’s robustness, with F1-micro and F1-macro scores of 94% and 93%, respectively. Results show that although women in STEM are underrepresented, 73% of the female alumni analyzed belonged to the short job waiting group. Furthermore, a strong negative correlation between GPA and job waiting time suggests that higher-GPA graduates tend to secure employment more quickly
Reputation-enhanced two-way hybrid algorithm for detecting attacks in WSN
Get PDFWireless sensor networks (WSNs) are susceptible to a variety of attacks, such as data tampering attacks, blackhole attacks, and grayhole attacks, that can affect the reliability of communication. We proposed a reputationenhanced two-way hybrid algorithm (RCHA) that uses cryptographic hash functions and reputation-based trust management to detect and de-escalate attacks accurately. The RCHA algorithm implements two hash functions RACE integrity primitives’ evaluation message digest (RIPEMD) and secure hash algorithm (SHA-3), to initiate the integrity check for the entire packet sent across the network. Every node in the WSN tracks a reputation score for each neighbor the node is connected to, and this score is dynamically updated based on the behavior of each neighbor. If a neighboring node’s reputation drops below a threshold, the node is sent a maliciousness designation. At that time, the node will broadcast an alert message to its neighboring nodes and begin to reroute its data through one of its trusted neighbors to ensure the reliability of the communication. The simulation results reported that the RCHA algorithm improved the accuracy of the attack detection rate and the number of packets delivered compared to traditional attack detection methods. The RCHA algorithm was able to maintain low computational and energy overhead for the WSN, making it an attractive option for a resource-constrained application in a WSN. Given the trends towards more collaborative networks, the reputation mechanism in the RCHA algorithm improves the overall reliability and capabilities of the WSN, regardless of adversaries
Fetal electrocardiogram extraction and signal quality assessment using statistical method
Get PDFAbdominal electrocardiogram (aECG) can be used to monitor fetal heart rate (fHR), providing critical insights into fetal health during pregnancy. However, separating the mixed signals of fetal ECG (fECG) and maternal ECG (mECG) within the aECG remains a critical challenge. This paper investigates the integration of statistical metrics, including signal-to-noise ratio (SNR), skewness, kurtosis, standard deviation, and variance to assess fECG signal quality during extraction using three adaptive filtering metods ((Least mean square (LMS), normalized LMS (NLMS), and recursive least square (RLS)) and independent component analysis (ICA). The findings reveal that RLS achieves the best performance among the three AF methods, with the highest SNR of 5.6 dB at the step size, µ of 0.9. For ICA with a bandpass Chebyshev filter (low-cut frequency = 1 Hz, high-cut frequency = 50 Hz) produces an SNR of 0.86 dB. Additionally, both RLS and ICA yield similar fHR values of 133 bpm with a PE measurement of 0.9%. In conclusion, integrating statistical metrics with ICA and RLS effectively extracts fECG with good signal quality. Future research could explore other ECG datasets and incorporate machine learning to further improve fECG extraction and signal quality assessment
A survey on fronthaul signaling of user-centric cell-free massive MIMO networks
Get PDFThe mandate for high data rates in mobile communication is increasing and will continue to do so in the future. Although the latest network technologies can meet this demand, they result in more-dense networks. Networks like ultra-dense networks and massive multiple-input multiple-output provide very high data rates, but they cannot meet the future demand. The main issue with existing networks is inter-cell interference and variations in quality of service esp. at the cell edges, leading to research on new network architectures that offer intelligent coordination and collaboration capabilities are being researched, like user-centric cell-free (UC-CF) massive-multipleinput-multiple-output (mMIMO). This network combines the best of ultradense networks and mMIMO and eliminates cell edge problems. It is served by access points that cooperate and coordinate with each other. This paper reviews the challenges and opportunities in physical layer parameterfronthaul signaling for UC-CF mMIMO. We discuss the basics of the network, the importance of fronthaul signaling, and propose various approaches in the literature to address challenges and identify research gaps and provide future directions. Our aims to provide a comprehensive overview of the current state of fronthaul signaling and highlight the key issues that need to be addressed to realize its full potential