International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
Performance Analysis of Task Distribution Mechanism in Multi-User’ Collaborative Assembly Task\u27 In 3d Virtual Environment
Get PDFCVEs are real-time, computer-simulated environments where two or more actors can mutually complete a task using synthetic objects. User performance is one of the major problems that arise due to coordination problems, a good mechanism to divide tasks, or less understanding or interaction among users collaborating. The impact of multi-user collaboration on using the task distribution mechanism remains unexplored. In this study, the impact of TDM on multi-users’ collaborative virtual environment is investigated. The TDM model assigns the task to collaborating users in CVEs on a static or dynamic manner. In static distribution, there exists weak coupling, and the amount of communication during the actual execution of a task is low, while in dynamic distribution, users are tightly coupled and hence need to communicate more. To study the effect of static and dynamic task distribution strategies on user’s performance in CVEs on multi users, a CVE prototype was developed using C++ and OpenGL, simulating an assembly task with distinct roles for multiple users, where twenty (20) group (each consists of two users) perform a task in collaboration under both strategies (static and dynamic) on two users and three users using arrow-casting and audio aids. The result shows that static with arrows-casting for two users takes an average time of 331.15 sec, and for three users, 321.45sec, and for audio (342.73sec and 326.34sec, respectively. Similarly, the dynamic with arrow casting for two users takes 347.76 sec, and for three users, 333.24 sec, and for audio, 350.12 sec and 344.4 sec, respectively. The findings provide valuable insights into how multi-user collaboration, task distribution methods, and cognitive aids can influence task efficiency and teamwork. However, when the number of users increased to three users, there is a chance that the performance will be degraded because, from the experimental data, a lower improvement was observed for three users than for two users. This research contributes to improving task management and collaboration in CVEs, with potential applications in training, education, and remote teamwork
Interactive Impacts of Heavy Metals and Soil Amendments on Enzymatic Activities and Microbial Biomass
Get PDFBoth organic and inorganic soil additives are frequently used to increase the bioavailability of lead (Pb) and cadmium (Cd) in polluted soils, but these amendments may also affect microbial activity in soils by modifying heavy metal solubility. This research assessed the influence of different soil additives on enzymatic activity and the solubility of Pb and Cd in spiked soils. Soils were spiked with Pb (0, 1000, 1500 mg kg⁻¹) and Cd (0, 100, 150 mg kg⁻¹) artificially. Incubation experiments were carried out with various amendments, such as citric acid (CA; 0, 10 mmol kg⁻¹), ammonium nitrate (AN; 0, 10 mmol kg⁻¹), EDTA (0, 5 mmol kg⁻¹), compost (CO; 0, 10%), and titanium dioxide nanoparticles (TNPs; 0, 100 mg kg⁻¹). The microbial biomass carbon (Cmic) and dehydrogenase activity (DHA) declined by 66% and 47% in Pb₁₅₀₀, and by 54% and 35% in Cd₁₅₀ treatments, respectively. In control soil, compost addition gave the highest value of Cmic and DHA, followed by TNPs, CA, AN, and EDTA. But the mixed application of Pb, Cd, and soil additives caused an overall reduction in microbial activity. Among all the treatments, EDTA alone and in combination with Pb and Cd showed maximum toxicity to soil microorganisms
The Comparative Study of Machine Learning Algorithms for Sentiment Analysis in Multimodal Medical Data: Comparative Study of Machine Learning Algorithms for Sentiment Analysis in Multimodal Medical Data
No full textSentiment analysis, a part of data mining, uses Natural Language Processing (NLP) to understand how people feel about certain topics or individuals. It focuses on the context and polarity of information, measuring public opinions from unstructured sources like social networks and healthcare websites. By extracting useful insights from this unstructured data, healthcare professionals can improve patient care, make accurate diagnoses, and provide personalized treatments. Machine learning (ML) plays a key role in this process. ML techniques like logistic regression, decision trees, and Naive Bayes have proven effective in tasks such as sentiment analysis and named entity recognition in medical data. The goal of ML is to create algorithms that enhance data processing and decision-making by identifying patterns that might be overlooked by humans. In this study, we compare the performance of three common ML models—(a) Logistic Regression, (b) Decision Tree, and (c) Naive Bayes—for sentiment analysis on medical image captions. The Radiology Objects in Context (ROCO) multimodal image and caption dataset was used for this NLP task. Caption pre-processing is done using filtering methods to improve text quality, followed by sentiment classification using pre-trained ML models. This comparison sheds light on the effectiveness of these algorithms in performing sentiment analysis in clinical settings
An AI-Powered Browser Extension Using Roberta and XAI for Phishing Email Detection and Security Awareness
Get PDFPhishing attacks are a common and serious cybersecurity threat today. They exploit human weaknesses by stealing sensitive information by sending fake emails and harmful links. Traditional email filtering systems like rule-based methods and black-box models, struggle to detect phishing. Rule-based filters fail when attackers use new tricks, and black-box models lack transparency, which limits user awareness.
This work introduces a smart browser extension that uses deep learning and Explainable AI (XAI) for phishing detection. We use a transformer-based model, Roberta, trained on a large email dataset, achieving 98.12% accuracy in classifying email content. For checking URLs, we use VirusTotal, which gathers threat intelligence from multiple sources. We also apply XAI tools to highlight key parts of the text that contributed to the classification of the email content, and a large language model (LLM) to provide simple explanations about phishing.
Our hybrid approach combines explainable deep learning with multi-source URL verification. This helps users understand phishing threats better and improves their ability to spot attacks on their own
Machine Learning-Based Fish Species Recommendation Using Water Quality Parameters
Get PDFThe integration of machine learning (ML) in aquaculture enables data-driven fish species recommendations based on water quality parameters. Traditional fish farming faces challenges like manual monitoring, inefficient species selection, and unpredictable water conditions, leading to economic losses. This paper presents a software-based fish recommendation system using ML models to analyze seven key water parameters—pH, Temperature, Turbidity, TDS, Dissolved Oxygen, Nitrate, and Ammonia. Various ML algorithms, including Random Forest, XGBoost, and SVM, were evaluated, with the optimized model achieving over 90% accuracy. A graphical user interface (GUI) allows users to input parameters and receive real-time recommendations, enhancing efficiency and sustainability in aquaculture
Cost-Effective Energy Management of a Microgrid Using a Hybrid Yellow Saddle Goatfish Optimization Algorithm
Get PDFThe increasing integration of renewable energy sources into hybrid Microgrid presents challenges such as power fluctuations, system complexity, and high operational costs. This paper proposes an optimized energy management framework that combines the Hybrid Yellow Saddle Goatfish Optimization Algorithm (HYSGA) with Sequential Quadratic Programming (SQP) to improve system efficiency, stability, and cost-effectiveness. The HYSGA approach efficiently manages energy distribution among solar photovoltaic (PV) systems, Battery Energy Storage Systems (BESS), and the power grid, ensuring reliable and cost-effective operation. HYSGA quickly identifies near-optimal solutions for complex energy management issues, while SQP fine-tunes these solutions to improve precision and convergence speed. Extensive simulations and cost comparisons confirm the framework\u27s performance. In the baseline scenario, the hybrid Microgrid incurs an annual operational cost of 13,800, achieving 49% savings. Further optimization with HYSGA reduces the cost to $13,430.08, resulting in a 50.118% savings. Additionally, comparative evaluations show that HYSGA outperforms traditional techniques like Mixed-Integer Nonlinear Programming (MINLP) in terms of cost savings, computational efficiency, and solution accuracy. This study provides a detailed analysis of the research methodology, solution approach, and performance evaluation, ensuring clarity. The results demonstrate that the HYSGA framework is a scalable, computationally efficient, and economically viable solution for hybrid Microgrid energy management. The proposed method offers a promising approach for enhancing energy efficiency and reducing costs in modern smart grid applications
Disease Detection Using Wrist Pulse Analysis
Get PDFEarly detection of diseases is crucial for effective treatment and management. Traditionally, disease detection involves invasive and costly medical procedures. However, recent advancements in non-invasive methods have proven highly successful in identifying various illnesses. The wrist pulse has long been an important tool for detecting diseases, with traditional Chinese medicine making extensive use of this method. It shows great promise in diagnosing a wide range of conditions. This study provides a detailed analysis of research on wrist pulse analysis and its applications in disease detection. It examines the physiological basis of wrist pulse analysis, focusing on the relationship between underlying medical conditions and the characteristics of wrist pulses. Additionally, the study explores how wearable pulse detectors and machine learning algorithms can improve the accuracy and effectiveness of wrist pulse analysis. In this research, we use a dataset of 300 samples from various diseases, analyzing it with MATLAB and applying ensemble learning algorithms. We have achieved accuracies above 80% for nearly all algorithms, and accuracy can be further improved by expanding the dataset with more samples and extracting additional features
Enhancing Predictive Business Process Monitoring in Call Centers through Multimodal Data Fusion and Heterogeneous Time-Aware LSTM-Based Multi-Task Learning
Get PDFThe optimization of call center operations and the enhancement of customer service are greatly supported by predictive business process monitoring. Traditional methods often overlook valuable multimodal data, such as conversations occurring in contact centers, because they typically rely on sequence data from business IT systems. This limitation hinders a complete understanding of business processes. In this study, we introduce a unique time-aware LSTM-based framework for predictive business process monitoring, which leverages both IT system data and dialogue data from contact centers. Our approach combines multiple data sources to improve the accuracy of forecasting ongoing business activities. To address challenges related to multi-task learning and to better utilize the rich information embedded in various data types, we propose a heterogeneous multi-task learning architecture called Heterogeneous Multi-gate Mixture-of-Experts (H-MMoE). Experimental results show that our method outperforms established baseline models such as Transformer, CNN, and standard LSTM. These findings demonstrate the potential of time-aware LSTM models to improve process monitoring, optimize workflows, and drive operational success in call center environments
AI-Powered Chatbot for Conversational Understanding in Roman Urdu
Get PDFMany people, especially in Pakistan and India, speak Urdu. However, when they write it online, they often use Roman Urdu (Urdu written with English letters). The problem is that most chatbots struggle to understand Roman Urdu because there is no standard way to write it—people spell the same words differently. This research aims to develop an intelligent AI chatbot that can understand and respond accurately in Roman Urdu. To achieve this, we will use advanced AI techniques such as Retrieval-Augmented Generation (RAG) and GPT-based models. The goal is to improve the chatbot’s accuracy and relevance, making it better at handling conversations in Roman Urdu. This study will explain how the chatbot is designed, trained, tested, and improved, helping AI work more effectively with languages that lack fixed writing rules
A Hybrid Approach to Fine-Grained Butterfly and Moth Classification Using Deep Features and Rhombus-Based HOG Descriptor
Get PDFButterfly and moth species are crucial for ecosystems as pollinators, pests, and biodiversity indicators, therefore necessitating their precise automated classification for extensive monitoring, conservation initiatives, and agricultural pest control. Nonetheless, considerable obstacles emerge from inter- and intra-species variety in wing coloration, patterns, posture, and the effects of lighting and background circumstances on pictures. This study presents a comprehensive framework that enhances feature representation via a dual-phase methodology. Initially, pictures undergo preprocessing by Contrast-Limited Adaptive Histogram Equalization (CLAHE) to augment distinguishing features. Subsequently, elevated semantic features are derived using a ResNet50 backbone pre-trained on ImageNet, with a baseline accuracy of 92%. A unique Corner Rhombus Shape HOG (CRSHOG) descriptor is suggested to accurately capture detailed geometric and textural wing properties, utilizing rhombus-based grid sampling and gradient orientation encoding. These complementary deep and handcrafted features are carefully integrated to form a hybrid representation, improving resilience to cluttered backdrops and position changes. The integrated feature set is assessed using several classifiers, with an Ensemble Subspace KNN model attaining the greatest classification accuracy of 94.6% on the Butterfly and Moth Image dataset, exceeding traditional CNN (Convolutional Neural Network)-only and HOG-based methods. These findings highlight the benefits of combining domain-specific shape descriptors with deep-learning features to enhance fine-grained insect categorization. Moreover, depending exclusively on standard RGB photos facilitates practical implementation on mobile and aerial platforms for real-time biodiversity surveillance and pest management. Future endeavors will concentrate on expanding this hybrid feature technique to encompass live video tracking and open-set species detection in uncontrolled settings