International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
High Isolation Low Profile MIMO Antenna for 5G Applications
Full text linkThis article focuses on the multi-input multi-output (MIMO) antenna with smaller dimensions and improved isolation between the ports which results in good performance and is suitable for modern wireless communication systems, especially USB dongle and 5G mobile applications. Two main problems with the conventional antennas are (i) Large height between the top to ground plate and (ii) Smaller isolation between the ports. These two problems are carefully addressed in the proposed design. The height in the proposed antenna is minimized up to 4mm between the top and to ground plate. Further achieved isolation is more than -18dB by introducing a novel type of periodic slots in the ground plate of the antenna. The antenna structure consists of a ground plate and two top plates, each with defined dimensions. The parameters are optimized using software for the best performance of the antenna to ensure efficient signal transmission and reception. The proposed antenna aims to explore the impact of dimension parameters like impedance matching and fractional bandwidth and isolation between the ports. The antenna resonates for 4.5GHz to 5.75GHz frequency bandwidth. The return loss for impedance bandwidth is below -20dB. The antenna is simulated on software and fabricated on substrate FR4. The simulated and measured results are compared which indicates that this antenna with high isolation and good impedance bandwidth is a more suitable candidate for 5G mobile applications. Compared to the conventional same-class antenna, the proposed antenna has good performance with minimized dimensions
Smart Gas Geyser with Real-Time Data Collection
Full text linkThis paper "Smart Gas Geyser with Real Time Data Collection" presents an IoT-based Smart Gas Geyser system designed to address energy inefficiency, safety risks, and user inconvenience in conventional water heaters. This paper proposes an IoT-based smart gas geyser system that enables remote monitoring and real-time data collection via a mobile application. This data-driven approach allows for automated adjustments and instant alerts, enhancing both safety and performance. The system integrates ESP32 with various sensors to monitor temperature, gas leakage, and flame presence. Data is transmitted to a Firebase real-time database, allowing users to make informed decisions via a mobile app. The user-friendly mobile app provides features such as temperature setting, real-time monitoring, and automatic shutoff, making geyser operation seamless and secure. The proposed system enhances energy efficiency, ensures safety, and provides cost-effective automation for domestic and industrial applications. The study also discusses experimental results, comparative analysis with conventional geysers, and future recommendations
Detecting Stance in Urdu Content on Social Media and Websites for Fake News and Propaganda Identification
Full text linkThe extensive spread of fake information has rendered various news types questionable, leading to a significant decline in trust in news. Social media is the primary channel by which fake news is disseminated widely. Worldwide, several deep learning methods have been created to identify fake news, achieving significant success with content in the English language. However, to our knowledge, there is no deep learning method available for detecting fake news or stance detection in content written in Urdu. Therefore, it is crucial to create a method that can detect fake news within Urdu language content. This study seeks to identify a method for detecting fake news in the Urdu language by proposing a framework that employs advanced Bidirectional Encoder Representations from Transformers (BERT), Embeddings from Language Models (ELMO), and various deep learning models (CNN, LSTM, Bi-LSTM) to evaluate performance accuracy on Urdu datasets (Liar-ProSOUL and Bend the Truth-Benchmark). We utilized Embeddings from Language Models (ELMO) for feature extraction and a convolutional neural network (CNN) for the classification task. The findings from the suggested framework indicate that ELMO excels with extensive dataset
NeuroSecure-IoMT: Deep Learning Meets Cyber Defense in the Internet of Medical Things
Full text linkIintrusion deduction systems (IDS) are crucial to preserving sensitive medical information from cyber threats. However, issues with multi-class intrusion detection include an imbalanced data set, poor accuracy for minority classes, and a lack of flexibility in handling complex real-world situations. To address these issues, we provide a hybrid framework that combines machine learning and deep learning methods to address these problems. The model uses a random forest classifier for anomaly detection after reducing dimensionality using an autoencoder. The Synthetic Minority Oversampling Technique (SMOTE) was used during processing to ensure equitable class representation and reduce class imbalance. A multi-class intrusion detection dataset tailored to healthcare applications was used to thoroughly test the suggested framework, which provides an impressive 99% accuracy rate. In addition to its excellent accuracy, the model addresses important issues in multi-class Intrusion detection by exhibiting remarkable precision for minority classes and consistent performance across all categories. These results highlight the framework\u27s effectiveness in providing dependable and effective normal detection solutions, which makes it ideal for implementation in crucial sectors like healthcare, their accuracy and data security are crucial
Hydraulic Modelling of Flood Assessment in Chenab Basin Pakistan
Full text linkFlooding represents one of the most recurrent and economically damaging natural hazards in Pakistan, particularly within the Indus River system, where extensive floodplains and dense human settlement exacerbate vulnerability. This study presents an integrated flood-assessment framework combining optical remote sensing, geographic information systems (GIS), and physically based hydraulic modelling to delineate and quantify flood inundation in the Chenab Basin, Punjab. Multi-temporal Landsat 8 OLI imagery acquired during pre-, peak-, and post-flood stages was processed to derive the Normalized Difference Water Index (NDWI), Modified NDWI (MNDWI), and Water Ratio Index (WRI) for flood detection. Comparative accuracy assessment using field observations and high-resolution reference imagery demonstrated that MNDWI outperformed other indices, achieving an overall accuracy of 91% compared to 83% for NDWI and 85% for WRI. Supervised maximum-likelihood land-use/land-cover (LULC) classification yielded an overall accuracy of 91.6% with a Kappa coefficient of 0.89, confirming strong agreement between classified outputs and ground reference data. A 30 m SRTM-derived Digital Terrain Model was employed to develop a one-dimensional hydraulic model in HEC-RAS, simulating flood scenarios for return periods ranging from 2.5 to 100 years (455,000–1,665,000 cusecs) along the Chenab River reach between Head Trimmu and Head Panjnad. Modelled water-surface elevations showed close correspondence with GPS-recorded flood marks, with positional deviations below 50 m and sensitivity analysis indicating a maximum ±0.15 m variation in water level for ±0.01 changes in Manning’s roughness coefficient. Results indicate that approximately 68% of the study area was inundated during the 2010 flood, with cropland accounting for nearly 61% of the affected area and settlements for 18%. The integration of satellite-derived water indices with hydraulic simulation proved effective for accurate flood delineation and hazard zoning, providing a robust and operationally scalable framework for flood-risk assessment and spatial planning in data-scarce river basins of Pakistan
Advanced Deep Learning-Based Potato Defect Identification Leveraging YOLOv8 for Smart Agriculture
Full text linkThis paper presents the design of an effective deep learning model to identify and rank potato defects, enabling intelligent farming and post-harvest tasks. The primary goal is to automate the quality measurement of potatoes in several categories: healthy, damaged, defective, fungal-diseased, and sprouted, with the help of an optimized YOLOv8 model. The data set on potato images was annotated and gathered in the real-world agricultural conditions in a wide variety of images. Data augmentation and transfer learning were used to train the model and enhance generalization and detection rates in different conditions. The experiment showed that the detection performance was high and it reached 95.3% training accuracy, 93.8% validation accuracy, and 92.5% test accuracy with an F1-score of 92.9. The results verify that the suggested approach plays a crucial role in detecting defects in potatoes in real time, which can be used to support comprehensive, computerized, and accurate agricultural surveillance
Leveraging Machine Learning for Spreading Factor Optimization in Lora WAN Networks
Full text linkThe Internet of Things (IoT) has witnessed exponential growth and widespread integration across diverse sectors such as agriculture, logistics, smart cities, and healthcare. Among various IoT communication paradigms, the Long-Range Wide Area Network has emerged as a prominent and preferred technology, attributed to its extended transmission range, energy efficiency, and cost-effectiveness. Nevertheless, the escalating proliferation of IoT endpoints has amplified the complexity of efficient resource orchestration, particularly in Spreading Factor (SF) optimization within infrastructures. To mitigate this challenge, this study introduces a Machine Learning–driven Adaptive Data Rate (ML-ADR) framework for dynamic SF management. A Long Short-Term Memory (LSTM) neural network was meticulously trained using a dataset synthesized via ns-3 network simulations to achieve optimal SF classification. The pre-trained LSTM model was subsequently deployed on end-device nodes to enable intelligent and adaptive SF allocation using real-time data during simulation. Experimental evaluations reveal significant enhancements in packet delivery ratio and notable reductions in energy consumption, thereby validating the efficacy and scalability of the proposed ML-ADR approach
Clear Tic-AI: Detection of Dysarthria and its Severity Analysis
Full text linkDysarthria and other motor speech disorders result from abnormalities in the neural or muscular processes that actually control speech production; conversely, these disorders affect the strength, coordination, and tone of the vocal muscles that ultimately produce less intelligible speech. Because dysarthria can range from moderate distortion of articulation to severe impairment of speech, early and accurate assessment is critical. The paper proposes Clearitic AI, an automated speech analysis platform that leverages artificial intelligence to diagnose vocal disorders. It fuses Wav2Vec2 with traditional acoustic features such as Mel-Frequency Cepstral Coefficients (MFCCs), pitch estimation, and spectral descriptors. Abnormal voice classification and its severity on a framework with a sequential neural network architecture are proposed. Extensive testing of the system is performed using 10,000 recordings of voice samples from the TORGO dataset and the Mozilla Common Voice dataset. Experimental results demonstrate that the proposed model achieves a classification accuracy of 94.2% (±1.3), an F1-score of 0.943, and an Area Under the Curve (AUC) of 0.987 on the test set, thereby establishing the effectiveness of this framework for dysarthric speech detection applications
Impact of Anticipatory Grief on Quality of Life among Caregivers of Thalassemia Patients: Mediating Role of Physical Activity
Full text linkCaregivers of patients with chronic illness, such as thalassemia, face unique challenges in their lives. During the caretaking of the patient, caregivers face anticipatory grief due to the fear of loss. The goal of the current study was to investigate the impact of anticipatory grief on the quality of life of caregivers of thalassemia patients with the mediating role of physical activity. Differences across genders in a proposed relationship were also studied. Anticipatory grief scale (Theut et al., 1991), international physical activity questionnaire (Geneva, 1998) & quality of life scale (Flanagan, 1970), alongside the demographic sheet and written informed consent form, were employed for collecting data among 300 caregivers using cross cross-sectional survey research design that included multiple cities of Punjab. Results indicate that anticipatory grief has a significant negative relationship with quality of life and physical activity. Physical activity has a significant positive relationship with quality of life. Results also indicate that women show significantly higher levels of anticipatory grief than men. Men show a higher level of quality of life than women. The findings of the study provided strong empirical support for the predicted role of anticipatory grief on quality of life among caregivers. These findings further highlight that physical activity should be part of parents\u27 lives to deal with grief related to the health of a child
Auscultation-Based Pulmonary Disease Detection and Classification Using Deep Neural Networks
Full text linkPulmonary diseases like Pneumonia, Bronchiectasis, and Chronic Obstructive Pulmonary Disease cause a large number of deaths worldwide. For such diseases to be treated and managed effectively, an early and accurate diagnosis is essential. In this work, we propose a deep learning model based on Recurrent Neural Networks (RNN) that can detect three different pulmonary diseases, as well as healthy lung sounds, using only auscultation recordings. The model was trained using the ICBHI dataset, which contains 920 recordings from 126 people and covers more than 6,800 respiratory cycles. To uniform the data, the audios are padded to equal length. To tackle class imbalance in the dataset, augmentation techniques of Gaussian noise injection, time-shifting, and time stretching are used. We employ a simplified version of the Gated Recurrent Unit (GRU)-based RNN architecture to deal with the padded sequences, along with a dropout layer to avoid overfitting. The model is trained using the Adamax optimizer with categorical cross-entropy loss, along with a model checkpoint to ensure learning consistency. Apart from the evaluation of model accuracy, we also evaluated the F1-score, accuracy, and loss graphs to ensure the competitive performance of our approach. Out of the six different experiments, with different data variations and two different model architectures, the outperforming model exhibited an accuracy of 98.53%, a precision of 98.57%, a recall of 98.53%, and an F1-score of 98.52%