Authors - Sivakami B U, M. Suresh Abstract - As the conversation shifts toward Industry 5.0 (I5.0), how prepared are manufacturing Micro Small Medium Enterprises (MSMEs) to adopt the foundational technologies of Industry 4.0 (I4.0)? Despite MSMEs' economic importance, they persistently face challenges in adopting essential technologies and undergoing digital transformation due to financial, infrastructural, and work-force-related barriers. The extant literature provides limited understanding of their readiness to implement individual I4.0 technologies. To address this gap, this study aims to develop a Fuzzy Logic-based readiness assessment framework to evaluate adoption potential and identify weaker attributes. Using an ontological approach, the framework identifies key enablers, criteria, and attributes. Findings reveal an “average ready” status, with critical gaps in IoT, Robotics, Automation, and Digital Twin technologies. The framework provides MSMEs with a roadmap to assess their readiness and strategically plan their I4.0 transition.
Authors - Drisya Murali, Suresh M Abstract - This study aims to develop a novel assessment framework that measures the practice of organizations in the construction industry for Construction 4.0. The research helps to measure and improve construction organizations’ practice level for Construction 4.0 by identifying key enablers and criteria and addressing weaker attributes. This research seeks to use a fuzzy logic approach within a conceptual framework to evaluate the current practice level of these organizations. By assessing organizational members' perspectives regarding strategic management, resource allocation, technological readiness, and socio-economic considerations, the study aims to identify critical practice criteria and categorize them into four enablers with detailed attributes. The assessment was done at one location, suggesting further evaluation could help improve practice levels across the organization. The management of a case construction organization would use the study to enhance the practice level of their organization to improve the Construction 4.0 practice. To enhance the organization's practice for Construction 4.0, focus on improving identified weaker attributes based on the proposed assessment framework model. The ultimate goal is to provide recommendations for improving practice level, thereby maximizing the benefits of Construction 4.0 in the construction industry. This study is the first to propose such a framework, contributing valuable insights to the research on Construction 4.0 in this sector.
Authors - Suhas Bhise, Devesh Kulkarni, Mayur Gaikwad, Parth Nevase Abstract - Warehouses and storage facilities are highly vulnerable to fires due to the presence of flammable materials and high-density storage layouts. Traditional fire detection and suppression systems frequently encounter limitations in such environments, such as delayed response times and restricted access to fire sources. The study describes the design and development of an automatic fire-fighting robot specifically for warehouses and storage facilities. The proposed robotic system combines advanced sensors for smoke, heat, and flame detection with real-time navigation capabilities to autonomously locate fire sources. Some of them uses thermal imaging cameras, infrared sensors, and LiDAR technology to precisely map the environment, detect obstacles, and navigate narrow aisles. The robot has a multi-directional water or foam nozzle for efficient fire suppression, as well as a decision-making algorithm that prioritizes critical fire zones, allowing for a quick and effective response. Furthermore, the system supports remote monitoring and control via a user-friendly interface, allowing for human intervention as needed. Heat-resistant materials and a long-lasting power supply improve the robot's durability, allowing it to operate continuously in extreme conditions. Experimental results show that the robot is effective at detecting fires early, responding quickly, and minimizing damage. The study seeks to provide a dependable, cost-effective solution for improving fire safety in warehouses and storage facilities, thereby reducing risk to both property and human life.
Authors - V. K. Abhang, Y. A. Shinde, S. N. Shingote, Adik Somal, Aglawe Nikhil, Akolkar Vivek, Ghondage Pratik Abstract - This project explores how machine learning can help optimize hybrid renewable energy systems, with a special focus on managing net metering costs. By analyzing real-time data from renewable sources and consumer energy usage, the goal is to create a smart, efficient framework that improves energy reliability while keeping costs low. The idea is to strike a balance ensuring that energy production and consumption align seamlessly with changing demand and environmental conditions. To make this happen, we’re using the Open Energy Modelling Framework (OEMOF), which helps optimize how energy is distributed, stored, and interacted with the grid. With OEMOF, we can simulate energy flows, make better decisions about energy trading and self-consumption, and develop cost-effective net metering strategies. On top of that, advanced predictive models for weather and energy demand forecasting allow for proactive system adjustments, making sure the setup remains efficient and reliable.Beyond the technical side, this approach directly supports the global shift toward sustainable energy. By making hybrid renewable energy systems more cost-effective and scalable, it not only helps individuals and businesses save money but also contributes to reducing carbon footprints moving us one step closer to a greener future.
Tuesday August 25, 2026 9:30am - 11:30am IST Virtual Room EGOA, India
Authors - Lokesh Khedekar, Suvarna Pawar Abstract - In order to increase the accuracy of threat detection for malicious URLs, this paper proposes an improved feature extraction methodology. High false positive rates are a common consequence of traditional detection systems' restricted feature sets. The suggested method extracts a wide variety of lexical, host-based, content-based, and character-level n-gram features in order to solve this. While host-based qualities offer contextual information like domain age and DNS validity, lexical features record structural irregularities. While n-gram features identify obfuscation through frequent character sequences, content-based features—such as login indicators, file extensions, and JavaScript references—reflect behavioral patterns. According to correlation analysis, the majority of traits are still weakly associated, providing a variety of complementary signals, even while some are strongly related. By properly recognizing benign URLs with 100% confidence, the suggested model showed dependable real-time prediction and obtained a high accuracy of 98.50% in detecting dangerous URLs.
Authors - Shreya Pulluri, Megha Mohan, Madduri Aditya Vardhan Reddy, Chitresh Simhadri, Geetha M Abstract - Bias in race and gender classification models creates strong ethical dilemmas, where disparities typically exist among groups of different demographics. In this research, a Bias-Aware Grad-CAM method is proposed that embeds interpretability-driven bias reduction within training. A ResNet model was trained for gender and race classification using the UTKFace dataset, which shows preliminary errors in minority group predictions. The Bias-Aware GradCAM framework proposed produces heatmaps to detect model attention areas, calculating Intersection over Union (IoU) between the heatmaps and pre-defined face bounding boxes. A reweighting process in the loss function, based on IoU scores, is introduced to retrain the model. Results show enhanced classification performance by aligning model attention with appropriate facial areas better. This strategy emphasizes the promise of explainability-guided methods to real-time bias reduction in face recognition models.
Authors - Swatee Nikam, Nilima Kulkarni, Amrita Manjrekar Abstract - With increasing data and its computation over clouds through various mediums it has become difficult to preserve privacy and then manipulate data according to the requirements. The third-party service providers have increased to do this manipulation over the bigdata which is collected over period of time. But again, the privacy is breached when the third party is involved with data of individuals and thus mechanism of homomorphic encryption (HE) is introduced which allows computation to be done on encrypted data. The initial work on homomorphic encryption were impractical but recent work is established with efficiency over practical application latest being embedded in Microsoft Edge browser. This paper gives a brief introduction to homomorphic encryption and its categorization over libraries created over recent years. Practical implementation HE is programmed over C++ language, with recent development the trial implementation is in progress over python language also. The libraries include such as Helib, SEAL, Blyss, concrete, etc. HE is developed as future enterprise for security and privacy guard up with application slated as geofencing with keeping secrecy of location or latigo polls for scheduling the meetings. The paper will discuss some other application with reference to new developed libraries under HE.
Authors - Mrunal Vibhute, Anjali Naik Abstract - In an era where misinformation spreads rapidly, the need for reliable and interpretable fake news detection systems is critical. This study evaluates deep learning models—Fully Connected Neural Networks (FCNN), Convolutional Neural Networks (CNN), Graph Convolutional Networks (GCN), Long Short-Term Memory (LSTM) networks, and FastText—for fake news classification. To ensure interpretability, Local Interpretable Model-agnostic Explanations (LIME) are applied to FCNN, CNN, and GCN, generating human-understandable explanations for predictions. While LSTM and FastText are included for performance comparison, they are excluded from interpretability analysis due to technical constraints. The models are trained using different input representations: TF-IDF for FCNN and CNN, and graph structures for GCN. This paper analyzes trade-offs between accuracy and explainability, offering insights into the effectiveness of deep learning models for fake news detection and contributing to the development of more transparent AI solutions.
Authors - Manan Parekh, Anmol Aafre, Parth Patel, Vaidehi Lad, Nirali Nanavati Abstract - The human eye is highly sensitive and vulnerable to various diseases that need timely attention; otherwise, they may lead to vision impairment. According to WHO (World Health Organization), most of these cases could be avoided through regular examinations. Therefore, early detection of eye disease has become a necessity. In this manuscript we propose a novel approach for prediction and classification of ocular diseases like age-related macular degeneration, cataract, diabetic retinopathy, glaucoma, hypertensive retinopathy, myopia, and normal using fundus images collected from multiple public sources. Our study comprises the study of multiple contemporary deep learning models, including EfficientNet-B4, ConvNeXt, DenseNet-201, and ViT-16B. Our proposed solution involves the implementation of the ConvNeXt model on our dataset, which achieves an accuracy of 94.14%. Our purpose behind this study is not only to develop an effective classification model but also to ensure its visual - understanding how and why the model is making certain decisions. To achieve this, we implemented the Gradient-Weighted Class Activation Mapping (GradCAM) to provide an explainable AI (XAI) based solution to help researchers and practitioners in the field.
Tuesday August 25, 2026 9:30am - 11:30am IST Virtual Room EGOA, India
Authors - Vinodh Kumar Minchula, Supraja Reddy A, Sharathchand Kodam Abstract - This paper investigates the characteristics of 5G NR-V2X channels and explores the benefits of relay-assisted communication for ensuring reliable and robust connectivity in dynamic vehicular environments. A detailed analysis of mmWave Base Station (BS) deployments utilizing MIMO antenna arrays of varying dimensions is conducted for the 5G NR n258 frequency band at 26 GHz. To evaluate system performance, a simulation framework is developed based on a Hyderabad Open Road Roundabout (ORR) scenario, aligned with 3GPP specifications, capturing vehicle mobility patterns and beam steering mechanisms to identify the optimal beam. To address the limitations of the NLoS, relay nodes are introduced to assist NLoS users, leading to measurable improvements in link quality and throughput. Experimental findings show that relay-assisted links yield up to a 4.1% increase in bit rate over direct NLoS connections. These results confirm the efficacy of relay-based architectures in enhancing signal robustness and extending mmWave communication range in V2X environments. The proposed framework provides a scalable and resilient solution for enabling high-performance vehicular networks in urban deployments. These findings clearly demonstrate that relay-assisted communication significantly enhances signal robustness and extends the effective range of mmWave links, thereby facilitating improved reliability and efficiency in next-generation vehicular communication networks.
Authors - Nirav Bhatt, Purvi Prajapati, Nikita Bhatt, Jiten Bhalavat Abstract - A crucial stage in medical image analysis for brain tumor diagnosis, treatment planning, and patient monitoring is brain tumor segmentation. It entails locating the tumor and any of its subregions, including the necrotic core, peritumoral edema and an enlarging tumor. Manual segmentation takes a lot of time and is prone to mistakes, which makes it unsuitable for regular clinical use. Recently, deep learning-based techniques have shown promise as a method for automatically segmenting brain tumors. In this work, we suggest a deep learning method for automatically segmenting brain tumors from magnetic resonance imaging (MRI) scans using a UNet architecture. A deep learning architecture created especially for image segmentation is the U-Net model. It is composed of an encoder-decoder structure, where the decoder reconstructs the input image and the encoder extracts features from it. On a range of medical image segmentation tasks, including brain tumor segmentation, the U-Net model has demonstrated state-of-the-art performance.
Authors - Divya Lakshmi R, Jayasri R, Deepak Gupta, Shobhana Palat Madhavan Abstract - The rise of social media has significantly influenced consumption behaviour, with influencers shaping purchasing decisions across industries. However, a countermovement— de-influencing—has emerged, urging consumers to rethink their buying habits and embrace conscious consumption. This study explores how de-influencers impact consumer decision-making by discouraging excessive and unsustainable purchases. This study employs the theoretical framework of Consumer Resistance Theory. Through an online survey of 192 respondents, the research examines factors such as trust in de-influencers, follower congruence, ethical concerns, and message content in shaping anti-consumption tendencies. The findings show that de-influencers are linked to attitudes that support sustainable consumption. There is a positive relationship between follower alignment and brand avoidance, while ethical concerns are strongly tied to anti-consumption and conscious consumption. A minimalist mindset also aligns with these patterns, indicating that de-influencers influence attitudes that lead to more sustainable choices. The study offers valuable insights for marketers, policymakers, and sustainability advocates, shedding light on the growing digital influence landscape and its implications for sustainable conscious consumption practices.
Authors - Pranali G.Chavhan, Ritesh V. Patil Abstract - The race between the rapid spread of ubiquitous computing and the Internet of Behavior has opened up a whole new avenue for the provision of personalized and context-aware services. To this end, the work presents a proactive recommendation algorithm that is meant to capitalize on IoB data to predict user behavior and deliver tailor-made content in an unobtrusive manner. With real-time behavior information added to the mix, it intends to go a step further from traditional recommendation systems. What differentiates this approach is its use and manipulation of a multitude of contextual factors: geographical context, temporal context, context of what device is being used - perhaps even emotional context as well. This live blending affords the system the ability to respond to what is happening in the real world, thus making it more reactive and relevant. The simulation results show that this awareness of context is going to generate a significantly better user engagement and accuracy of recommendation rather than traditional systems.
Authors - D Kishore Babu, K Subba Rao, Nagesh Babu Dasari, Kumara Raja, Golla Mary Prakash Kumari, Vijayakumar Chilamkurthi Abstract - A neurological disorder identified as Parkinson's disease (PD) is described through a continuing loss of dopamine producing brain cells, which results in bradykinesia, rigidity, and tremors. Symptoms typically appear after 60 - 80% of these cells have disappeared. 7 - 10 thousand people suffer with Parkinson's disease throughout the world, primarily affecting those over fifty, while 4% of cases also affect younger age groups. 90% of Parkinson's disease patients experience speech issues early in the disease. Machine learning algorithms offer a practical means of diagnosing Parkinson's disease early on by analyzing speech features. By using voice datasets from the UCI Machine Learning Repository, these methods may effectively and with low error rates classify Parkinson's disease (PD). This raises the likelihood of an early diagnosis and course of action.
Authors - Manasi Sangamnerkar, Prachi Mukherji, Seema Rajput, Nandini Kendre, Vaishnavi Mudaliar Abstract - This paper gives a comparative study of the K-means clustering algorithm run on three platforms: a CPU, an FPGA, and a hybrid CPU-FPGA setup, focusing on execution efficiency and scalability. The CPU version is suitable for small datasets due to its simple serial processing ability, while the FPGA shows superior performance for larger datasets with hardware acceleration and parallel processing. The hybrid setup employs the ARM Cortex-A9 processor in addition to the programmable logic of the Xilinx ZedBoard (ZYNQ-7000 SoC). The algorithm is run through Vitis on the CPU, while AXI-interfaced IP cores, developed using Vivado, provide signal monitoring and real-time debugging through the Integrated Logic Analyzer (ILA). This setup provides dynamic software control and high-speed processing. The FPGA showed an execution time of 38.077 nanoseconds, compared to the 0.015519 seconds on the CPU, providing a speedup of about 106 times. Implementation issues, such as the lack of native floating-point support and reliance on fixed-point approximations, have been noted for future improvement. Additionally, 8-bit binary representations of centroids are visualized using LEDs on the FPGA, providing a physical and intuitive visualization of the clustering process. This paper illustrates the effectiveness of FPGAs and hybrid CPU-FPGA setups in accelerating compute-intensive machine learning algorithms and the benefits of hardware-based optimization in real-time and embedded systems.
Authors - Roshni De, Debatosh Chakraborty, Dwijen Rudrapal, Baby Bhattacharya Abstract - Floods are one of the most dangerous natural disasters, both frequent and dynamic due to continuous land use changes and climate change. This causes difficulty in predicting the areas most vulnerable due to their complex nature, causing heavy loss and damage. The study presents a data-driven framework for flood susceptibility mapping, examining the influence of multiple satellite-derived geo-spatial and temporal features in the Cachar district of Assam, India—a region frequently impacted by monsoonal flooding. By integrating Machine Learning with features derived from NDVI (Landsat 8), LULC (Sentinel-2), topographic variables (SRTM DEM), soil texture (OpenLandMap), and monsoon precipitation (CHIRPS), alongside flood extent information obtained from NDWI and Sentinel-1 SAR data, the model aims to enhance predictive accuracy in flood-prone, data-constrained environments. A rigorous feature selection process using IGR and VIF score and comparative evaluation across various classifiers was used to optimize the model. The study highlights the importance of integrating machine learning with remote sensing data to construct a precise flood risk model to aid the disaster management team in identifying vulnerable regions.
Authors - Shri Harini A, Niranjana Shaji, Karthick rajaa A S, Gugapriya G Abstract - Visual impairment increases fall risk, particularly among older adults with low vision facing a 16% higher likelihood of falls and those with blindness experiencing a 40% increased risk. To address this, the research presents an ML-driven, IoT-enabled smart cane equipped with sensor-based behavior analysis for real-time fall detection and mobility assistance. The system analyzes motion patterns and sudden orientation changes that helps in detecting falls while integrating obstacle detection with multi-modal feedback. Designed with low-power and cost-effective embedded components, the system ensures efficiency on resource-constrained devices, while IoT connectivity enables remote monitoring and caregiver communication. This smart cane offers a costeffective, scalable solution to improve independence and quality of life for visually impaired individuals.
Authors - Subhashree Banerjee, Ranit Roy, Anirban Chattopadhyay Abstract - Having utilized the latest scientific and knowledge developments, Wireless-Sensing node Technology has made many strides in healthcare today. Many people, however, are made to suffer from different health-related issues and even death due to several illnesses, often from an absence of proper medical attention. There is an urgent need for an efficient, modern real-time patient monitoring system realized through the power of IoT. Continuous real-time tracking of vital health parameters, like temperature, blood pressure, oxygen saturation measurement, and electrocardiogram (ECG), and displaying their values other than ECG on LCD, along with alerts issued in case of deviations or abnormality for the reported health factor to the concerned healthcare professional, and storing the real-time stats of the patient in graphical form, makes up the proposed innovative system. Incorporated with different specialized sensors, including the temperature sensor, blood pressure sensor, SPO2 sensor, and ECG sensor, and also providing two microcontrollers accompanied by software, the system is intended to meet the primary purpose of establishing a robust patient management infrastructure rooted in IoT. By adopting this high-tech system, the health workers will be better placed to monitor their patients anywhere they might be, either in a hospital setup or even at the convenience of their homes, through an integrated IoT-enabled healthcare platform. The overarching goal of this initiative is to guarantee the delivery of top-tier patient care services while promoting enhanced health outcomes and overall well-being for individuals under medical supervision.
Authors - Jyotsna More, Suvarna Aranjo, Martina D’souza, Siddhi Awlegaonkar, Saahil Chaurasia, Aditya Ghadge, Shreya Jadhav Abstract - Traditional voting systems suffer from several challenges, including voter impersonation, ballot tampering, multiple voting, and lack of transparency, which compromise electoral integrity. To address these issues, this paper proposes a Blockchain-Based Biometric Voting System that integrates fingerprint authentication with blockchain technology to enhance the security, transparency, and reliability of elections. Biometric authentication ensures that only registered voters and administrators can access the system, eliminating impersonation and fraudulent voting. The R307 fingerprint module is used for real-time voter authentication, preventing unauthorized access. Once verified, votes are recorded on a blockchain ledger, ensuring data immutability and decentralization. Unlike conventional databases, blockchain technology eliminates single points of failure, preventing vote manipulation and unauthorized modifications. The system also incorporates a web-based interface that allows voters to register, authenticate, and securely cast their votes, while administrators can manage the election process with full transparency. Blockchain’s decentralized nature ensures that all transactions, including candidate registration, vote counting, and election results, are securely stored and verifiable, preventing tampering and external interference. Initial testing demonstrates high accuracy in biometric verification and efficient blockchain-based vote storage, making the system scalable for local, state, and national elections. By integrating biometric security with blockchain's trustless nature, this system provides a fraud-resistant, transparent, and tamper-proof voting solution, fostering greater public confidence in electoral processes.
Authors - Mathesh H P, Shanthini E, Praveen S, Praneshwaran M S Abstract - Environment with green practices depends heavily on efficient waste management, especially in parts of the economy where garbage is produced and processed in huge quantity Plastic bottles are one of the largest impediments to waste material due to their bulk use and destructive effect on the environment Inability of conventional methods of garbage segregation to offer the required precision and speed to function in most situations renders garbage segregation ineffective. To counter the issue, in this project, an improved waste sorting mechanism using the YOLOv8 object detection algorithm is utilized. Image processing and real-time machine learning are utilized by the system to separate and filter plastic bottles from the rest of the waste materials on a conveyor belt accurately. On grounds of efficiency and effectiveness, the YOLOv8 algorithm is superior to traditional methods and previous models. It is also highly renowned for detecting objects with a very high degree of accuracy. Other methods did not extend beyond detection, but auto-sorting ensures that plastic trash is sorted according to what needs to be recycled. This raises the overall processes of waste management and lowers contamination. Because it can provide businesses with a more and more scalable option, this invention is a giant leap for automated waste management.
Authors - Kaushal Kotkar, Samiran Deore, Riddhi Tak, Tejal Deshmukh, Rupali Vairagade, Nilakshi Jain Abstract - Traditional CAPTCHAs often hinder users more than they stop bots. This project proposes a passive, user-friendly alternative that monitors behavior—like mouse movement, typing speed, and clicks—to distinguish humans from bots. Built with Python, FastAPI, MongoDB, and XGBoost, the system defends against threats like DoS/DDoS attacks while remaining seamless. It adapts over time through model updates and achieved 95.3% accuracy with minimal false positives. With response times under a second, it outperforms conventional CAPTCHAs in both speed and usability.
Authors - Samrudhi Pustole, Hemal Rajput, Linisha Thakor, Dhanashri Gawai, R. B. Murumkar Abstract - Fraud detection is a critical challenge in financial transactions, requiring advanced machine learning models to distinguish between genuine and fraudulent activities. This project focuses on LSTMbased fraud detection, leveraging historical transaction data to identify suspicious patterns. The model processes multiple attributes, including transaction amount, category, user job type, geolocation, and time-based parameters, to assess fraud risk. In addition to the LSTM model, we conducted single-attribute fraud analysis using various models, evaluating their individual impact on fraud detection. This helped determine the most influential features in predicting fraudulent transactions. The system is integrated into a web-based application built with React and Flask, allowing users to input transaction details and receive a fraud score in real-time. The backend ensures efficient data preprocessing using feature scaling and categorical encoding, aligning new transactions with the trained model’s feature space. Through extensive testing with high-risk and low-risk transaction scenarios, the system demonstrates its ability to detect fraudulent transactions with high accuracy, making it a valuable tool for financial security. . . .
Authors - Shirina Samreen, K. Shailaja Abstract - The goal of this research is to accurately classify Android applications as either malware or legitimate software using machine learning techniques. This accuracy is attained through a well-organized approach for dimensionality reduction utilizing an Autoencoder to transform a high-dimensional feature space to a compact representative feature space. This approach is crucial for reducing dimensionality, especially since the novel NATICUSdroid dataset used for Android malware classification contains a large number of features, including both native and custom permissions. The primary contribution of the research is the design of a Machine Learning Pipeline that prioritizes the most relevant features, ensuring high accuracy with a minimal set of features. Classification is performed using various ensemble classifiers. Predictive ability of the proposed MLP is assessed through various evaluation metrics using a confusion matrix.
Authors - Akshat Sharma, Alka Chaudhary Abstract - This study aims to develop an Automated Incident Response System (AIRS) for real-time detection and mitigation of cybersecurity threats. The system employs Random Forest for anomaly detection using live network traffic data. SMOTE is applied to address class imbalance, and Optuna optimizes model parameters for enhanced accuracy. A web-based dashboard provides real-time visualization of security incidents. Performance evaluation on the UNSW-NB15 dataset demonstrates high accuracy (94.7%) and reduced false positives, ensuring robust cyber defense.
Tuesday August 25, 2026 3:30pm - 5:30pm IST Virtual Room EGOA, India
Authors - Swati Suman, Sumit Ray, Ajay Kumar Prusty, Umesha C, Girish Prasad Rath, Sabyasachi Patnaik, Ankita Priyadarshini, Swagat Shubhadarshi, Pavan Kumar Pandey, Lalithamma M Abstract - Climate change has a significant influence on agriculture, affecting developing nations' food security and financial condition. Thus, the use of Digital Twins, Internet of Things (IoT) devices, and Artificial Intelligence (AI) may play an important role in transforming agriculture that is data-enabled in real time for crop development, high productivity, or climate mitigation. These technologies would aid in predicting drought start periods, optimizing irrigation scheduling to react to any specific climatic shift, and driving crop rotations in a given area. To power climate-resilient farming development, AI and IoT must be combined, resulting in DTs. This technology incorporates agricultural offices, animal monitoring, crop harvests, crop protection, and a DT for predictive maintenance purposes. AI is transforming agriculture by analyzing large volumes of data to forecast climate change consequences. Precision agriculture, a key AI tool, uses micro-localized applications based on syntactic sensory data, drones, and satellite data. Smart agriculture uses IoT, AI, Big Data analytics, and DTs to gather, integrate, and analyze data from various sources. AI-powered models can forecast future weather patterns, insect infestations, and disease outbreaks, enabling earlier intervention and higher output. These insights enable improved resource allocation, agricultural practice optimization, and enhanced farm output in the face of climate change and hence making the DT the possible game changer in the field of agriculture while keeping sustainability as one of its important cornerstones.
Authors - Sreelasya Changalasetty, Lalitha Saroja Thota, Seshagiri Rao Kandukuri, Suresh Babu Changalasetty, Ahmed Said Badawy, Wade Ghribi, Sajid Ali Khan, Syed Asif Basha, Firdouse Banu Abstract - Child-related criminal offenses are amongst the utmost heinous, targeting the vulnerable youth of the society. These crimes include physical & sexual abuse of children, child labor, child trafficking, cyberbullying etc. The World Health Organization (WHO) estimates that up to 1 billion crimes against children have occurred globally. In India alone, more than 350 such crimes are reported each day. This study focuses on identifying crime hotspots related to children in India using hierarchical clustering, a machine learning technique. The research utilized crime data from the National Crime Records Bureau (NCRB) India for 2016–2020, alongside state-wise child population estimates, to group states according to the gravity of offenses committed against children. The data was pre-processed, normalized, and analyzed using KNIME software, which applied a bottom-up hierarchical clustering approach to create a dendrogram for visualizing crime clusters. The results revealed three category Indian states of crime zones in India: high, medium, and low. Delhi state was identified as the primary hotspot with a very high crime rate, followed by 17 states in the medium-risk category, and 12 states in the low-risk category. These findings underscore the need for targeted interventions and enhanced child protection policies, especially in Delhi. The study illustrates how hierarchical clustering can be effectively applied to criminology for identifying high-risk areas and informing policy decisions. Future research may include the use of localized data and exploration of other clustering algorithms to refine and improve crime analysis and prevention strategies.
Authors - G. B. Sambare, Gauri Pawar, Soham Vhanamane, Tejas Sonar, Omkar Gouroji Abstract - Phishing, a deceptive practice aimed at acquiring sensitive information through fraudulent websites mimicking legitimate ones, remains a significant cybersecurity threat. This paper presents a survey of machine learning (ML) algorithms applied for the detection of phishing URLs. We explore various feature categories derived from URL structure, domain characteristics, and HTML/JavaScript content. In particular, the characteristics of interest involve address-bar-based features (i.e., URL length, redirection patterns, existence of IP addresses), domain-based features (e.g., DNS records, website age, web traffic), and HTML/JavaScript-based features (e.g., iframe redirects, disabling the right click). The accuracy of a variety of classification methods, namely Decision Tree, Random Forest, XGBoost, and Support Vector Machines (SVM), is covered. By our results as well as available literature, we point out the effectiveness of XGBoost, which, in our testing, reached a comparatively high value of 86.8% accuracy and exhibited its prowess as a solid detector of phishing URLs. This paper offers some vision into the benefits and shortcomings different machine learning approaches are for phishing assaults.
Authors - Ajay K.Talele, Gayatri Bhurguda, Siddhant Yenpure, Purva Ratnaparkhi, Jidnya Santosh Jadhav, Rugvedi Nimbhore, Advait Mhalungekar, Tanishka Kalokhe, Riddhi Rathi, Roshan Raut, Pratha Sawant Abstract - In an era where skill development is essential for career growth, learners often struggle to track progress, find curated resources, and stay motivated. SkillTrax is a personalized learning tracker built to address this gap. The platform helps users enter their skills, select proficiency levels, set learning goals, and track their progress with recommended resources and quizzes. It is powered by object-oriented design principles, ensuring modularity and scalability. This paper outlines the motivation, design, and implementation details of SkillTrax, focusing on its core features, backend architecture, and educational impact.
Tuesday August 25, 2026 3:30pm - 5:30pm IST Virtual Room EGOA, India
Authors - Owais Ansari, Hemangini Patel, Tejas Maroo, Morvi Panchal, Nikita Raichada Abstract - In recent years, emotional voice conversion and expressive speech synthesis have gained attention due to their applications in areas such as automated dubbing, human-computer interaction, and assistive technologies. Our research proposes an AI-based dubbing system, SyncVox, which presents a seamless voice dubbing custom pipeline designed to provide seamless voice conversion across languages. It addresses low-resource video dubbing using various advanced technologies like speech recognition, translation, and style transfer. The pipeline combines speaker embeddings with advanced techniques to produce natural-sounding, speaker-alike voice synthesis. By employing multitask learning with Text-To-Speech, the pipeline is capable of capturing rich linguistic information while retaining languages; this allows content creators to dub videos without compromising the original speaker’s intent and naturalness. Early results show that this system effectively synthesizes natural-sounding speech with high emotional fidelity.
Authors - Sunitha Ratnakaram, Venkamaraju Chakravaram, Chakravaram Sri Surya Narayan Raj Abstract - Artificial Intelligence (AI) has significantly reshaped the landscape of Information and Communication Technology (ICT), particularly in the corporate sector. Integrating AI into programming and ICT development has led to automation, enhanced efficiency, and optimized business operations. AI-driven solutions are not only transforming software development but also impacting key business functions such as cybersecurity, finance, marketing, and decision-making processes. AI-powered automation has enabled businesses to achieve unprecedented productivity levels, making operations faster and more accurate while reducing human intervention. Companies now rely on AI for predictive analytics, customer insights, fraud detection, and even real-time strategic decision-making. This research paper explores the transformative role of AI in ICT programming, highlighting its vast impact on corporate efficiency, software development methodologies, cybersecurity frameworks, and business functions such as marketing and finance. It also delves into the ethical considerations of AI implementation, the challenges posed by AI-driven ICT automation, and the potential of AI to reshape the global economy. This study presents various case studies and empirical findings that illustrate how organizations have successfully incorporated AI into their ICT programming strategies to gain competitive advantages. Furthermore, the research investigates future trends, exploring how AI is expected to evolve within the corporate sector and the potential risks it may pose. The researcher used descriptive exploratory research methodology. The findings of this paper contribute to the ongoing discourse on AI's role in shaping the digital landscape, offering insights into both opportunities and concerns surrounding AI adoption in the modern corporate world.
