Daily Archives: April 7, 2026

Authors: Dadarkar Ehaan Mubasshir, Ansari Ayan Atif Masood Iqbal, Khan Zain MD Irfan

Abstract: MentorAI is a comprehensive, web-based intelligent learning assistant designed to transform how students study, retain knowledge, and engage with educational content. The platform integrates a personalized AI Tutor powered by large language models, a Voice Recall system for active retrieval practice using the Web Speech API, an SM-2 algorithm-driven Spaced Repetition Flashcard engine, an AI-generated adaptive Quiz Engine, an AI- assisted rich-text Workspace for notes and PDF imports, a 3D Knowledge Graph for visual concept mapping using D3.js force-directed visualization, and a command-palette-style Nexus navigation system. A central Dashboard aggregates learning metrics, study streaks, mastery scores, and AI-detected weak topics in real time. This paper presents the complete system architecture, feature design rationale, technology stack, database design, security model, testing methodology, and results achieved during the development of MentorAI as a final year engineering capstone project.

Published by: vikaspatanker

Authors: Sparsh S. Misal, Yash R. Lodha, Parth P. Gargote, Shivam S. Daundkar

Abstract: Road infrastructure plays a critical role in transportation, but issues like potholes significantly affect safety, efficiency, and maintenance costs. Traditional pothole detection methods rely heavily on manual inspection and public reporting, which are often delayed and inefficient. This project proposes a smart pothole detection and reporting system using computer vision and machine learning. The system uses a live webcam feed to detect potholes in real-time using a model trained with Teachable Machine and deployed using TensorFlow.js. When a pothole is detected, the system captures an image, records the location, date, and time, and automatically generates a complaint ticket. The backend, built using Flask, stores the report data and provides a history of detected potholes. This system offers a low-cost, scalable, and automated solution that can be extended for smart city applications and real-time road monitoring systems.

Published by: vikaspatanker

Authors: Dheeraj Patil, Sanika Dixit, Aditya Dixit, Meenakshi Deotare

Abstract: Air pollution is one of the most serious environmental threats in urban areas, affecting both human health and climate. Traditional air quality monitoring systems provide only point-based information; hence, this limits their ability to show distributions across a city. Herein, this work describes HEAL, a web-based system for pollution hotspot predictions and visualizations through the utilization of machine learning and data visualization techniques. This system collects air quality data from APIs or sensors, processes it, and generates dynamic heat maps that showcase the levels of pollution in real time. Interpreting the interaction among environmental, traffic, and meteorological data, HEAL offers citizens, policymakers, and researchers new localized insights into air quality variations, which will result in better decision-making.

DOI:

 

 

Published by: vikaspatanker

Authors: Dr. A.Avinash, Kanchi Dhanusha, Saladi Rudra Naga Prasanna Lakshmi, Thiguti Sri Ajitesh, Kesanakurthi Satya Karthikeeyan, Vangapandu Lokesh

Abstract: Thyroid disease is one of the most common endocrine disorders affecting millions of people worldwide. The thyroid gland plays a crucial role in regulating metabolism, growth, and overall body functions. Any imbalance in thyroid hormone production can lead to conditions such as hypothyroidism or hyperthyroidism. Early detection of thyroid disorders is important to prevent serious health complications and to ensure timely treatment. Traditional methods of diagnosing thyroid disease rely on laboratory tests and manual evaluation, which may be time-consuming and sometimes prone to errors. With the advancement of artificial intelligence, deep learning techniques can assist medical professionals in improving diagnostic accuracy and reducing workload. In this project, a deep learning-based Multi-Layer Recursive Neural Network (ML-RNN) is proposed for thyroid disease detection and classification. The system includes data preprocessing, feature selection using the Fisher Score method, and classification using the ML-RNN model. The dataset used for analysis is obtained from a standard repository and includes various thyroid-related attributes. The performance of the proposed model is evaluated using metrics such as accuracy, recall, precision, and error rate. Experimental results show that the ML-RNN model achieves better performance compared to traditional machine learning algorithms such as Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and Random Forest (RF). The proposed approach provides an effective and reliable method for thyroid disease detection.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.155

Published by: vikaspatanker

Authors: Dr. Y. Jayababu, Gokeda Veera Satya Sri Pravallika, Ayyarapu Teja, Appasani Hari Kailash Chowdary, Addanki Yuva Sai Surya Prakash, Garapati Poorna Venkata Ranjit Kumar

Abstract: The study suggests an accessible and secure machine learning model for forecasting brake failures in large commercial vehicles. We support this proposal with evidence. Heavy transport vehicles' Air Pressure System (APS) is constantly monitored by IoT-based sensors in modern day heavy transport systems, generating vast amounts of operational data. Detecting brake failures manually with large and highly unbalanced datasets is time-consuming and inefficient. Our approach to these problems involves the use of K-Nearest Neighbour (KNN) imputation for missing values and SMOTE for dealing with class imbalance. Both methods are effective in both situations. Logistic Regression, Decision Tree, Support Vector Machine, Gradient Boosting, and Random Forest are among the machine learning algorithms that undergo stratified cross-validation during implementation and evaluation. The Random Forest classifier's accuracy, precision, recall, F1-score and ROC-AUC are shown to be more than satisfactory using experimental data. Enhanced transparency and trust in the prediction process are achieved through the use of Explainable Artificial Intelligence (XAI) techniques like SHAP and LIME, which can interpret model decisions. They also use methods of selecting features that reduce computational complexity while preserving high levels of accuracy in making predictions. This proposed framework improves fault detection reliability, reduces maintenance costs and allows for predictive maintenance in heavy transport systems.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.154

Published by: vikaspatanker

Authors: Mr.M.Raja Kumar, Pepakayala Bhavani Sri Alekhya, Dasari Asritha, Sada Uma Maheswara Rao, Thimmasatthi Venkateswarlu, Kollu Rajesh

Abstract: The rapid growth of Internet of Things (IoT) devices has significantly improved automation, connectivity, and data-driven decision-making across various domains such as healthcare, smart cities, agriculture, and industrial systems. However, the increasing number of interconnected devices has also introduced serious security challenges. IoT-enabled cyber-physical systems are highly vulnerable to cyber-attacks such as Distributed Denial of Service (DDoS), data injection, botnet attacks, and unauthorized access. Traditional machine learning techniques often struggle to provide high detection accuracy due to imbalanced datasets, high-dimensional features, and inefficient parameter tuning. In this project, a hybrid deep learning-based intrusion detection framework is proposed for identifying security attacks in IoT-enabled cyber-physical systems. The proposed model combines Convolutional Neural Network (CNN) and Deep Belief Network (DBN) to improve feature learning and classification performance. To enhance the model’s efficiency and convergence speed, a novel hybrid optimization technique called Seagull Adopted Elephant Herding Optimization (SAEHO) is employed for tuning the classifier weights. The proposed framework is evaluated using standard IoT intrusion detection datasets such as UNSW-NB15 and BoT-IoT. Performance is measured using metrics including accuracy, precision, sensitivity, specificity, False Positive Rate (FPR), False Negative Rate (FNR), and Matthews Correlation Coefficient (MCC). Experimental results demonstrate that the hybrid classifier optimized using SAEHO outperforms conventional machine learning and optimization-based models in terms of detection accuracy and reduced error rates. The developed system provides an effective and scalable solution for enhancing security in IoT-enabled cyber-physical environments.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.153

Published by: vikaspatanker

Authors: Mrs.KanakaTulasi P.Reddi, Sai Varshitha Kuppili, Gabu Ganesh Sasikanth, Adapa Sai Teja Venkata Vinay, Medaboyina Karthik, Trivinesh Gundra

Abstract: Autonomous marine vessels rely heavily on artificial intelligence systems for accurate sea state estimation, which plays a crucial role in navigation, stability control, and operational safety. However, AI-based models are vulnerable to adversarial attacks, where small and carefully crafted perturbations in input data can significantly degrade model performance. Such attacks may compromise safety and reliability, especially in critical maritime environments.This project proposes a novel robustness-enhancing adversarial defence approach to improve the reliability of AI-powered sea state estimation systems. The framework focuses on strengthening deep learning models against adversarial perturbations while maintaining high estimation accuracy. The system integrates adversarial training and defensive mechanisms to enhance model stability under uncertain and hostile conditions. Experimental evaluation demonstrates that the proposed defence strategy significantly improves robustness without sacrificing predictive performance. The results confirm that the enhanced model maintains reliable sea state estimation even in the presence of adversarial disturbances.The proposed approach contributes to improving the safety, security, and reliability of autonomous marine navigation systems.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.152

Published by: vikaspatanker

Authors: Mrs.M.Uma Devi, Togaru Reshma Sri, Katikidala Satya Ratna Naveen, Gubbala Leela Madhavi, Kalvakolanu Venkata Pavan Chaitanya, Velduti Srivenkata Surya Sai Kumar

Abstract: The rapid advancement of generative artificial intelligence has made it increasingly difficult to distinguish between real images and AI-generated synthetic images. Modern diffusion models can produce highly realistic visuals that closely resemble authentic photographs, raising serious concerns about misinformation, digital fraud, and media manipulation. As synthetic image generation becomes more accessible, reliable detection mechanisms are essential to maintain digital trust and security.This project presents an image classification framework for identifying AI-generated synthetic images using deep learning techniques. A balanced dataset is constructed by combining real images from the CIFAR-10 dataset with synthetic images generated using Stable Diffusion. A Convolutional Neural Network (CNN) model is trained to perform binary classification, distinguishing between real and fake images. In addition to classification, Explainable Artificial Intelligence (XAI) techniques such as Grad-CAM are applied to interpret model decisions and visualize the regions that influence predictions.Experimental results demonstrate that the proposed model achieves high accuracy in detecting synthetic images while maintaining reliable generalization performance. The explainability component further enhances transparency by revealing distinctive patterns and artifacts present in AI-generated images. The proposed system contributes to improving digital image forensics and strengthening defences against AI-driven visual misinformation.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.151

Published by: vikaspatanker

Authors: Dr. Y V Ram Kumar, Keerthi Teja Sri, Patamsetti Yaga Sri Satya Prasanna, Chintapalli Sashank, Theripogu John, Surapureddy Venkata Sai Praneeth

Abstract: Accurately estimating traffic states is a crucial aspect of transportation engineering, enabling effective traffic control and operations. In recent years, Physics-Regulated Deep Learning (PRDL) has gained significant attention due to its ability to achieve higher accuracy while requiring less training data compared to conventional deep learning (DL) approaches. However, a key challenge of PRDL is the lengthy training time required for closely related but distinct tasks.To address this limitation, this paper introduces a hybrid physics-regulated deep transfer learning approach that leverages the strengths of transfer learning, PRDL, and DL to enhance estimation accuracy and reduce computational costs, particularly in scenarios with limited observation data. The proposed framework includes two transfer learning variants designed to extract and transfer essential features from pre-trained models to new but similar traffic environments. This hybrid approach integrates deep learning training, minimizing computational overhead by eliminating physics-based loss calculations during training.Simulation results demonstrate that, compared to traditional PRDL methods, the proposed transfer learning approaches improve estimation accuracy by over 12% on average while reducing training time by more than 50% on average. These findings highlight the potential of hybrid transfer learning techniques in accelerating the adoption of PRDL for traffic state estimation, making it a valuable tool for transportation systems with limited computational resources.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.150

Published by: vikaspatanker

Authors: Dr. K. Mounika, Rachakonda Shashikanth, Palivela Prem Chandu, Virothula Sasi Kumar, Balusu Eswar, Kandikonda Pranay

Abstract: The accurate forecasting of Daily Global Solar Radiation (DGSR) is a vital tool in renewable energy planning, climate research, and environmental monitoring. This paper presents a proposal for utilizing machine learning to enhance the estimation of DGSR by using satellite image data. This method utilizes reflectance values obtained from Metaset Second Generation (MSG) satellite images across multiple spectral channels and relies on ground-based meteorological parameters rather than traditional models. Artificial Neural Network (ANN) and Support Vector Machine (SVM) are two supervised machine learning regression models that are utilized for forecasting solar radiation. How do they compare and contrast? The Gharda radiometric station in Algeria has collected measurable solar radiation data for four years (2014-2017) using inputs from satellite imagery, which are then combined to create the dataset. To evaluate these models, statistical performance metrics such as Root Mean Square Error (RMSE), Normalized RMSEA (NRMSE) and MAE (Made Absolute Percentage Error), MBE (Merck-McGregor), and correlation coefficient (R) are utilized. Prediction accuracy is significantly influenced by the number and combination of satellite input parameters, as demonstrated by experimental data. Compared to the SVM, the ANN model had a better RMSE of 21221. The NRMSE, MAPE, and MBE have all been reported with 3.46%, 2.85%, 7.26, etc, respectively. Wh/m2. A 0.99 correlation coefficient is associated with a Wh/m2 value.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue2.149

Published by: vikaspatanker