International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
Photocatalytic Degradation of Deltamethrin in Drinking Water Under Visible Light by Using Zno and Tio2
Get PDFThe use of deltamethrin is increasing due to its high demand in agriculture. However, it is toxic to both surface and groundwater. Agriculture plays a crucial role in the economy of any major nation. This study aims to enhance pesticide degradation by using specially designed catalysts optimized for visible light exposure. The key innovation lies in the customized catalyst design, which improves photocatalytic efficiency while offering a cost-effective and environmentally friendly approach. Various factors affecting degradation, including adsorbent quantity, pH, contact time, and initial concentration, were analyzed. The reactor consists of a 6-watt (380 nm) visible light lamp and a stirrer to ensure uniform mixing of the sample. Photocatalysts ZnO and TiO₂, in concentrations ranging from 0.1 to 3.0 g/L, were used to generate oxidizing agents. Under visible light, the impact of these factors on the degradation of different pesticide solutions was examined. The optimal doses were found to be 1.5 g/L for ZnO and 0.1 g/L for TiO₂. ZnO achieved a degradation rate of 96.3%, while TiO₂ slightly outperformed it with a rate of 96.34%. The study also investigated the effect of pH variations on deltamethrin degradation, revealing stronger degradation in alkaline conditions. Additionally, TiO₂ effectively reduced the COD value, demonstrating its superior efficiency in pesticide breakdown
Addressing Class Imbalance in Credit Card Fraud Detection: A Hybrid Deep Learning Approach
Get PDFThe rise of credit card fraud is a global concern, demanding reliable detection methods that can overcome challenges with imbalanced datasets and limited exploration of hybrid modeling approaches. This study introduces a hybrid deep learning architecture combining Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) layers alongside SMOTE-TOMEK preprocessing to address imbalanced data issues in credit card fraud detection. The research analyzes a substantial dataset containing both legitimate and fraudulent transactions, evaluating the performance of GRU, LSTM, and the novel Hybrid model through comprehensive data exploration, preprocessing, and feature selection. Performance evaluation uses metrics including accuracy, precision, recall, F1 Score, AUROC, and AUPRC. The experimental results demonstrate the effectiveness of deep learning architectures, with AUROC values of 0.974551 for LSTM, 0.958174 for GRU, and 0.976205 for the Hybrid model. The Hybrid model showed particularly promising results with a precision of 0.9121 and AUPRC of 0.886068, outperforming the individual models. These findings indicate that combining complementary deep learning architectures enhances fraud detection by leveraging their respective strengths in capturing both long-term dependencies and transaction patterns. These insights offer valuable guidance to financial institutions in implementing effective fraud detection systems while emphasizing the importance of continuous improvement of deep learning algorithms to address evolving cyber threats
VDMF: VANETs Detection Mechanism Using Fog Computing for Collusion and Sybil Attacks
Get PDFVehicular Ad Hoc Networks (VANETs) have evolved as a key component of the intelligent transportation system, enhancing road safety and traffic efficiency. It is crucial to secure sensitive information, and detection of incident response, whenever malicious activity is observed. Key components of VANETs include vehicles, Roadside Units (RSUs), and Fog servers (FS). Despite this, the open and evolving nature of VANETs introduces substantial security challenges, including exposure to malicious attacks like Sybil and collusion attacks. The proposed technique addresses the crucial security vulnerabilities in VANETs by developing a robust and efficient fog computing-based mechanism for detecting and mitigating Sybil and collusion attacks. The proposed approach emphasizes minimizing computational and communication overheads while ensuring timely and accurate detection and response to malicious activities. The results show that the proposed technique provides less communication and computational overheads in sparse and dense scenarios with enhanced security
Low Aperture High Gain Antenna for Wi-Fi Applications
Get PDFThis article proposes a design for a low-aperture, high-gain antenna specifically tailored for Wi-Fi applications. This project aims to enhance the performance of Wi-Fi networks by increasing signal strength and coverage area. Conventional dielectric rod antennas face three main challenges: first, they typically have low gain and excessive length; second, they exhibit high side lobe levels; and third, as the antenna length increases, side lobe levels also rise alongside gain. The proposed antenna structure effectively addresses these issues. The novelty of the proposed antenna lies in its ability to achieve greater gain at the same length compared to conventional antennas while maintaining low side lobe levels that do not increase with antenna length. The proposed design features a Yagi-Uda configuration on a printed circuit board (PCB) made from FR4 epoxy with a dielectric constant of 4.4, combined with a tapering dielectric Teflon rod with a dielectric constant of 2.1. The antenna was simulated using HFSS software, fabricated, and then tested to compare simulated and experimental results, which indicate that the proposed structure primarily operates at a frequency of 5 GHz. It achieves performance within the frequency band of 4.8–5.3 GHz, with a fractional bandwidth of 10%. At these frequencies, the structure provides a directivity of 16.4 dBi. A comparison of results demonstrates that the presented antenna outperforms traditional antennas in the same class, making it suitable for Wi-Fi, WLAN, and satellite applications. This revision enhances clarity, coherence, and overall readability while preserving the original intent and details
Parametrical Analysis of Symmetrical Double U-Slots Micro Strip Circular Patch Antenna for Wireless Communication Devices
Get PDFThe advancement in telecommunication demands antennas having wide bandwidth, compact size, and high performance. On the other hand, the microstrip antenna has a narrow bandwidth and low radiation efficiency. Recently, a number of researches were made to improve the bandwidth of microstrip antenna. The proposed antenna is based on to studied parametrically the microstrip antenna and their effect has been analyzed in terms of return loss, radiation patterns, and current distribution on the surface of the patch. The parameters such as the radius of the patch, lengths and widths of the slots, and feed point location are changed to different values using CST Microwave Studio. In this article double U-slotted in circular patch microstrip antenna are designed. The circular patch of radius 18mm and thickness of 1mm, FR4 substrate of permittivity of 4.3, dimension of (40 X 50) mm2, the thickness of 3.6mm, and coaxial probe are used to design the proposed antenna. The designed antenna has resonance frequencies at 3.2GHz, 5.8GHz, and 6GHz and the simulated gains are 3.72dBi, 7.67dBi, and 8.02dBi respectively. The VSWR at the resonance frequencies 3.2GHz, 5.7GHz, 5.8GHz, 5.94GHz, 6GHz, and 6.14GHz are 1.50, 1.38, 1.42, 1.60, 1.34 and 1.61. The VSWR is less than 2 at all resonance frequencies which shows good impedance matching. The return loss at the resonance frequencies of 5.8GHz and 3.24GHz is -19.04dB and -16.38dB respectively. The antenna has a bandwidth of 0.68GHz ranging from 5.52GHz to 6.18GHz. The proposed antenna is suitable for many wireless applications such as WLAN (5.15 – 5.35 & 5.75 – 5.8) GHz, Wi-Fi (5.15 – 5.82) GHz, and RFID (5.725 – 5.875) GHz
Modified Convolutional Neural Networks for Facial Emotion Classification
Get PDFFacial expression analysis is a fascinating yet challenging problem in the realm of artificial intelligence. The vast variability in human expressions poses a significant hurdle for machine learning methods to detect them accurately. Recently, machine learning and deep learning approaches have made notable strides in this area, leveraging Deep Neural Networks (DNNs) to identify human emotions. Convolutional Neural Networks (CNNs), in particular, have proven effective in resolving the complexities involved in human facial expressions, making them a preferred choice for these tasks. In this study, we proposed a modified CNN architecture by introducing a new layer to enhance accuracy. The CNN network is trained on both frontal face images and images with varying poses. We utilized three distinct datasets FER 2013, CK+ and our own dataset to achieve the desired results. The evaluation results obtained using the proposed network surpass those achieved by conventional CNN networks. Notably, our proposed network achieves an average accuracy of 97.5% on our collected dataset
Advanced AI Mechanics in Unity 3D for Immersive Gameplay: A Study on Finite State Machines & Artificial Intelligence
Get PDFThis research explores the history and operationalization of cutting-edge AI technologies, developed for Unity 3D video engine, in particular how Artificial Intelligence (AI), animation, and FSMs have been used in video games. AI looks at the various elements incorporated in the game design as a means of augmenting the player experience with a focus totally on a game, where the player controls a Paladin character, alongside an array of enemy characters, including skeletons and mutant bosses. The second aspect provides a look at why it is necessary to experiment with such elements from the gamification perspective, with a specific interest in how player experience can be prolonged with these components. It also aims to design these components so that provide players the satisfaction of playing the game, engagement after every session, and most importantly encouraging the players to come back and play the same scenario over and over again. In addition, this dissertation considers the principles and practice of the game in detail assessing how it fits in the category of an immersive experience. We have focused on how integrating FSMs and animation aids in creating smart actions in gaming characters to expound on the interactivity of the player. Because of these observations, AI will be a key factor in the transformation of future game development, which is boosted to raise the current level of gaming by inspiring more impressiveness and repeatability of gameplay
An Identification of Fake Contents Using Text-mining Techniques
No full textIn recent years, social media users have become increasingly concerned about sharing content that may be unpleasant or harmful. The widespread use of platforms like Facebook and Twitter has contributed significantly to this growing awareness. The primary objective of our approach is to accelerate and automate the detection of offensive content posted on these platforms, simplifying the process of taking necessary actions and filtering harmful communications. A benchmark dataset, OLID 2019 (Offensive Language Identification Dataset), is available online to aid in this task. Our study focuses on identifying whether a tweet is offensive. Our team, which included several members, rigorously compared various feature extraction methods and model-building algorithms. Ultimately, our comparative analysis revealed that decision trees were the most effective model. The decision trees applied to the normalized dataset resulted in an 84% improvement in the Macro F1 score, which aligns with previous research. In conclusion, a real-time system could be developed across multiple social media platforms to detect and evaluate objectionable posts, enabling timely interventions to promote healthier online behavior and foster a positive societal impact
A Large Language Model based Web Application for Contextual Document Conversation: A Large Language Model based Web Application for Contextual Document Conversation
No full textThe emergence of LLMs, such as ChatGPT, Gemini, and Claude has ushered in a new era of natural language processing, enabling rich textual interactions with computers. However, despite the capabilities of these new language models, they face significant challenges when queried on recent information or private data not included in the model’s dataset. Retrieval Augmented Generation (RAG) overcame the problems mentioned earlier by augmenting user queries with relevant context from a user-provided document(s), thus grounding the model’s response to inaccurate source material. In research, RAG enables users to engage interactively with their documents, instead of manually reading through their document(s). Users provide their document(s) to the system, which is then converted into vector indices, and used to inject contextual information into the user prompt during retrieval. The augmented prompt then enables the language model to contextually answer user queries. The research is composed of a web application, with an intuitive interface for interacting with the LIama 3.2 1B, an open-source LLM. Users can upload their document(s) and chat with the LLM in the context of their uploaded document(s)
Bitcoin Price Forecasting: A Comparative Study of Machine Learning, Statistical and Deep Learning Models
Get PDFIntroduction/Importance of Study:
Cryptocurrency price prediction is crucial for investors and researchers, given the market\u27s nonlinear nature and the potential for significant financial implications.
Novelty:
This study offers a novel approach to cryptocurrency price prediction, leveraging a range of machine learning and deep learning models to address the challenges of predicting Bitcoin\u27s exchange rate.
Materials & Methods:
The study employs various machine learning and deep learning models, including Recurrent Neural Networks (RNNs) such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), along with traditional models like Linear Regressor, Random Regressor, ExtraTreesClassifier, XGBoost Regressor, ARIMA, Prophet, and CNN.
Results & Discussion:
The ExtraTreesClassifier model emerged as the top performer, achieving a Test MAPE of 0.0689. This model outperformed deep learning models like RNNs, indicating its effectiveness in cryptocurrency price prediction.
Conclusion:
The findings suggest that the proposed models, particularly the ExtraTreesClassifier, can provide valuable insights for investors and traders in the cryptocurrency market