Authors: Mr.MD. Abdul kala, V.Krupa, M.pavani M. Hemanth sai

Abstract: Diabetic Retinopathy (DR) is a serious eye disease caused by long-term diabetes. It is one of the main causes of blindness around the globe. Early detection and prompt treatment are crucial to prevent permanent vision loss. Unfortunately, traditional diagnostic methods depend on the manual inspection of retinal fundus images by ophthalmologists. This process is time-consuming, subjective, and requires specialized skills. This project presents a Hybrid CNN-MLP Model for automated detection and classification of diabetic retinopathy using retinal images. The system combines Convolutional Neural Networks (CNN) for feature extraction and Multilayer Perceptron (MLP) for classification. The CNN component effectively captures spatial features like microaneurysms, hemorrhages, and exudates. Meanwhile, the MLP classifies these features into different levels of DR severity. The system is created using Python, TensorFlow/Keras, and Flask for online interaction. Users can upload retinal images, enter patient information, and receive real-time predictions with confidence scores, medical suggestions, and downloadable PDF reports. The system also keeps a record of patient history and provides visual analytics through graphs. This proposed model shows better accuracy, efficiency, and usability. It serves as a valuable tool for early screening and supports healthcare professionals in making decisions.

DOI: https://doi.org/10.5281/zenodo.19549841

 

Published by: vikaspatanker

Authors: Mrs. G. Monika, B. Bindu, U. Edukondalu, K.Varshith

Abstract: Insider threat is one of the biggest problems facing organizational security since insiders are individuals with authorized access to an organization’s information assets. Organizational security solutions can only detect outsider attacks and do not perform effectively when faced with malicious behaviors or accidental acts carried out by insiders. In this research paper, a method of detecting insider threat using behavioral anomaly is outlined. This solution aims at continuous observation of user behavior such as logging on, file access and general interaction with the system resources. Machine learning algorithms are employed in modeling user behavior and alerting any deviation that can imply an act of malice.

DOI: https://doi.org/10.5281/zenodo.19549305

 

Published by: vikaspatanker

Authors: Rajat Vijaybahadur Singh, Prasad Balasaheb Varpe, Pranav Navnath Lokhande, Krushna Suresh Awate

 

Abstract: Agriculture plays a vital role in the economy, and efficient farming techniques are essential for increasing crop productivity. Fertilizer application is one of the most important processes in agriculture, but traditional methods of applying fertilizers are time-consuming, labor-intensive, and often result in uneven distribution. To overcome these problems, a solar-powered fertilizer spraying machine is developed in this project. The main objective of this project is to design and fabricate a cost-effective, eco-friendly, and efficient fertilizer spraying system that reduces manual effort and ensures uniform spraying. The machine consists of a solar panel, battery, solar charge controller, water motor pump, storage tank, nozzle, flow pipes, and a four-wheel frame. The solar panel converts sunlight into electrical energy, which is stored in the battery and used to operate the motor pump. The pump creates pressure to spray the fertilizer solution through the nozzle in the form of fine droplets. The system provides several advantages such as reduced labor, time saving, uniform distribution of fertilizers, and low operating cost due to the use of solar energy. The four-wheel structure makes the machine portable and easy to operate in agricultural fields. It is especially useful for small and medium-scale farmers. This project demonstrates the effective use of renewable energy in agriculture and contributes to sustainable farming practices. The developed machine is simple in design, economical, and capable of improving overall agricultural efficiency and productivity.

DOI: https://doi.org/10.5281/zenodo.19548929

 

Published by: vikaspatanker

Authors: P. Dr. K. Rama Linga Reddy, Penumarty Srilakshmi Bhanupriya, Nikitha Mora, B. Suchitra, Sruthi Gujjula

Abstract: The paper describes the design of an Autonomous Signal Deception and Offensive System to be used at the Battlefield – a Simulink-based RF electronic warfare (EW) simulation system based on an earlier created hardware prototype, the Ultrasonic Deception System. In the previous system, four deception methods, including range deception, angle deception, stealth, and noise injection are shown with Arduino Mega and ultrasonic sensors, and the choice of the technique is done manually by an operator. The proposed system completely removes any manual involvement and adds five important extensions: autonomous selection of deception techniques with a randomized decision engine, an offensive electromagnetic pulse (EMP) generation subsystem, a cryptographic Identification Friend or Foe (IFF) protocol based on challenge-response authentication via XOR operations and pre-shared secret keys, an accurate RF channel model including path loss, propagation delay, and additive white Gaussian noise (AWGN). This system is implemerandi-based technique selection, and a Countermeasure Generation block that generates high-amplitude EMP pulses. The results of simulations show that autonomous threat classification is successful, the deployment of unpredictable deception techniques, and the possibility to quantify the degradation of the enemy system. Performance is analyzed based on six metrics such as IFF classification accuracy, deception effectiveness, SNR degradation, Shannon channel capacity, deception unpredictability entropy and system health degradation rate.

DOI: http://doi.org/

Published by: vikaspatanker

Authors: Dr. P.Vamsi krishna raja, Nama Venkata Bhaskara Sudheer

Abstract: Electric load forecasting plays a crucial role in efficient power system operation and energy management. Accurate prediction of electricity demand helps in reducing operational costs and improving system reliability. However, traditional forecasting methods often fail to handle complex and non-linear patterns present in real-world data. To address this issue, this paper proposes a machine learning–based approach using Extreme Gradient Boosting (XGBoost) for electric load forecasting. The proposed system utilizes historical load data along with important features such as time and temperature to train the model. Data preprocessing and feature selection techniques are applied to improve data quality and model performance. XGBoost, a powerful ensemble learning algorithm, is employed to capture complex relationships and enhance prediction accuracy. The model is evaluated using standard performance metrics, and the results demonstrate improved accuracy and efficiency compared to conventional methods. The proposed approach provides a reliable and scalable solution for electric load forecasting, supporting better decision-making in power system planning and management.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.179

Published by: vikaspatanker

Authors: K V Bharadwaj Karthik, V. Abhilash Naik, V Kowshik Chavali, G Suresh Babu

Abstract: This project focuses on the analysis of a three-phase star–delta step-down transformer using Altair Flux with emphasis on the no-load test and short- circuit test. The objective is to accurately evaluate core (iron) losses and copper (Joule) losses through finite element electromagnetic simulation. In the no-load test, rated voltage is applied to the primary winding while the secondary is kept open, enabling determination of magnetizing current, flux distribution, and core losses. The Bertotti loss model is employed within Altair Flux to separate hysteresis, eddy current, and excess losses in the core. Flux density distribution is examined to ensure operation below saturation limits. In the short-circuit test, the secondary winding is shorted and a reduced voltage is applied to circulate rated current, allowing evaluation of winding resistance, leakage reactance, and copper losses. The simulation accurately captures current density and I²R losses in both primary and secondary windings. The star–delta connection is properly modeled to obtain correct phase relationships and loss values. Results from Altair Flux demonstrate realistic loss estimation consistent with transformer theory. The study confirms that core losses remain nearly constant with load, while copper losses vary with the square of current. Overall, the work validates Altair Flux as an effective tool for detailed electromagnetic analysis of transformer performance using standard no-load and short-circuit test procedures.

Published by: vikaspatanker

Authors: Mr. Shashank Tiwari, Kommunuri Ashok Kumar, Valigonda Laxmiprasanna, Kayitha Sai Rachana

Abstract: The Conversational Product Recommendation System using LLM is an AI-driven application designed to enhance product recommendation by enabling natural language interaction between users and the system. In modern e-commerce environments, users often face information overload due to the vast number of available products. Traditional recommendation systems rely on static filtering methods and fail to understand complex user queries expressed in natural language. To address these limitations, the proposed system integrates Large Language Models (LLMs) with Natural Language Processing (NLP) techniques to interpret user intent, preferences, and constraints. The system provides a chatbot-based interface where users can interact conversationally, refine their queries, and receive personalized product recommendations in real time. A recommendation engine processes extracted features and ranks products based on relevance, while a backend database manages product data and user interactions.

Published by: vikaspatanker

Authors: M. Sujana Priyadarshini, Akula Swathi

Abstract: The exponential growth of email communication has led to an increase in unsolicited and potentially harmful spam messages, posing significant challenges to both users and organizations. Traditional spam detection techniques primarily focus on classification accuracy while often neglecting data security and privacy concerns. This paper presents a secure and efficient email spam detection system that integrates machine learning with cryptographic techniques. The proposed approach utilizes Support Vector Machine (SVM) for effective classification of emails based on textual features. To ensure data confidentiality, Advanced Encryption Standard (AES) is employed for encrypting email content, while Elliptic Curve Cryptography (ECC) is used for secure key exchange. The integration of classification and encryption mechanisms enables the system to provide reliable spam detection while preserving sensitive information. The proposed framework is suitable for real-world applications where both accuracy and data privacy are essential.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.175

Published by: vikaspatanker

Authors: M. Sujana Priyadarshini, Gunduprolu vijayakumar

Abstract: The study proposes an intelligent and reliable hybrid framework for enhancing fishermen safety and communication in marine environments using machine learning and electromagnetic water networks. Fishing activities in deep-sea regions involve significant risks due to unpredictable weather conditions, accidental border crossings, and limited communication facilities. Traditional monitoring systems rely heavily on manual observation and basic GPS tracking, which are often inefficient in handling real-time emergencies and dynamic ocean conditions. Additionally, the lack of continuous monitoring and predictive capabilities increases the vulnerability of fishermen to accidents and environmental hazards.To address these challenges, the proposed system integrates real-time data acquisition from multiple sources, including GPS tracking, environmental sensors, and electromagnetic sensors, to ensure continuous monitoring of marine conditions. The system employs machine learning techniques such as anomaly detection algorithms to identify abnormal vessel behavior, including sudden stops, unusual movements, and route deviations that may indicate distress situations. Furthermore, time-series data collected from sensors is analyzed using advanced deep learning techniques to predict environmental changes such as weather fluctuations and sea conditions.The model is trained and evaluated to accurately detect potential risks and provide early warning alerts, thereby enabling proactive decision-making. The proposed multi-layer framework enhances system performance by combining real-time monitoring, anomaly detection, and predictive analysis. This integrated approach improves communication between fishermen and coastal authorities through wireless technologies, ensuring timely response during emergencies.The system significantly enhances maritime safety, reduces the risk of accidents, and improves operational efficiency. By leveraging machine learning and real-time data processing, the proposed solution provides a scalable, efficient, and intelligent framework for ensuring the safety and security of fishermen in modern maritime environments.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.176

Published by: vikaspatanker

Authors: P. Chakradhar Rao, Vakadi venkata krishna

Abstract: The study proposes an efficient and secure hybrid framework for detecting Android malware in modern mobile environments. The widespread adoption of Android smartphones has led to increased security risks, as these devices are frequently targeted by sophisticated malware attacks. Furthermore, the growing integration of Android applications with Internet of Things (IoT) systems amplifies the potential impact of such threats. Detecting malware manually in large-scale and continuously evolving datasets is both time-consuming and ineffective. To address these challenges, our approach integrates real-time data acquisition and deep learning techniques. Malware hash values are dynamically updated using data extracted from Twitter at regular intervals of 48 hours, ensuring the system remains up-to-date with emerging threats. In addition, application features, particularly permissions, are analyzed using a Recurrent Neural Network (RNN) with Long Short-Term Memory (LSTM) architecture for accurate classification. The model is trained and evaluated to distinguish between benign and malicious applications, achieving a detection accuracy of approximately 94%. The proposed multi-layer framework enhances detection efficiency by combining traditional signature-based methods with intelligent learning mechanisms. This integrated system improves reliability, strengthens mobile security, and provides an effective solution for real-time Android malware detection and prevention.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.177

Published by: vikaspatanker