Authors - Sanchit Prashant Joshi, Vedant Vipin Joshi, Aditya Arun Mangalekar, G.S.Mundada Abstract - Malware classification is essential in cyber-security. It en ables prevention of threats by identifying and accurately classifying ma licious software. It also helps in understanding attacker behavior, enhanc ing threat intelligence, and improving the overall effectiveness of security systems. It is increasingly critical as adversaries now employ obfuscation techniques to avoid detection. Traditional models such as Convolutional Neural Networks (CNN) often struggle with such obfuscated malware samples. In this paper, we propose MalViT, a Vision Transformer (ViT) based framework for robust malware classification using grayscale image representations of malware binaries. The ViT is fine-tuned on a prepro cessed Malimg dataset. To evaluate the robustness of the model, real world obfuscation techniques such as Encryption, Dead code insertion, Random masking and Junk Padding are simulated. ViT model is initially f ine-tuned on the clean samples and later on a combination of the clean and obfuscated samples. Both models are evaluated on the clean and obfuscated test sets to highlight the robustness of the model. The final model achieved a combined accuracy of 94.52 % on both the clean and obfuscated samples. The results demonstrate that MalViT maintains a competitive performance under obfuscation. This project highlights the potential of ViTs in building resilient malware classification systems and provides a foundation for future work in transformer based architecture for malware analysis.
Authors - Samiksha Ganesh Zagade, Arya Mahesh Parkar, Suman Madan Abstract - Advances in Artificial Intelligence, Machine Learning and Internet of Things technologies have enabled wearable devices to sense as well as process and respond to human behaviour in real time. While most wearable devices today are used for health and fitness tracking. Many people face communication challenges such as language barriers, difficulty understanding emotions or social cues, social anxiety and accessibility issues for individuals with hearing or speech impairments. Existing systems often collect data but fail to provide meaningful, real-time assistance during actual human interactions. This research paper presents a literature-based study on AI powered wearable devices designed to support and enhance human communication. The research papers are focusing on intelligent wearables that use multimodal sensors such as microphones, cameras and sensors. These systems apply AI techniques to interpret speech, gestures, facial expressions and emotional signals in real time. The wearable devices considered include everyday consumer-oriented systems such as smart eyewear that provides audio visual assistance and wrist worn wearables that offer haptic feedback. The key focus of this study is to examine how such devices can deliver subtle, real-time support through visual prompts, audio cues or vibrations to improve conversational awareness and user confidence. The expected outcome is to identify current capabilities, practical limitations and design considerations for developing human centric wearable technologies that move beyond passive tracking toward meaningful communication support.
Authors - Adnan Hasan, Ishaan Mishra, Jyotiska Bose, Jada Viswa Chaitanya Sai, Jai Kumar, Kaif Akhter, Ranjita Kumari Dash Abstract - In the present-day context, presentations and computer-based interac tion play a crucial role in various domains, particularly in education and business. Traditionally, users have to rely on physical devices such as mouses, keyboards, or laser. Although these devices meet the basic requirements, they still reveal many limitations regarding mobility, continuity, and dependence on battery life. To address these limitations, hand gesture-based presentation control systems have emerged as a promising solution due to their intuitive, natural, and engaging interaction style. This paper proposes a touchless system that enables users to control common desktop operations as well as presentations in a natural manner using hand gestures captured via a standard webcam. The proposed system lev erages OpenCV for real-time video acquisition and preprocessing, while Medi aPipe framework is employed for hand tracking and landmark extraction. From the experiments, our system can process in real-time with the accuracy of approx imately 92%. As a result, users can seamlessly control slides, use virtual mouse operations, annotate presentation content, and engage with the audience in a more interactive and natural way without physical contact.
Authors - Jyoti Chandel, Meenakshi Mittal Abstract - Internet of Things (IoT) devices are growing in domains because of their reliability and efficiency in monitoring, real-time detection and automated support. However, these IoT systems have also introduced security challenges. These devices are vulnerable to cyber threats, where attackers exploit weak points in the system to steal sensitive information. One of the attacks is the Distributed Denial of Service (DDoS) attack, which disrupts services by overwhelming systems and making them inaccessible to legitimate users. IoT devices are resource-constrained, so reducing feature dimensionality is essential to lower computational overhead and complexity. IoT devices generate data for detecting cyber-attacks, but sharing such data across organizations raises privacy concerns. To address these challenges, the proposed approach is designed in two phases. In the first phase, a hybrid feature selection technique using mutual information, permutation feature importance, and Greedy wrapper-based feature selection with cross-validation is applied to extract relevant features. In the second phase, Federated Learning (FL) is applied to train the model without sharing raw data among clients. Within the FL framework, Random Forest (RF) algorithm is utilized for training due to its robustness and classification capability. The proposed model is evaluated under two data distribution scenarios: mildly non-IID and strongly non-IID conditions. Experimental results demonstrate that the model achieved an accuracy of 99.69% in a mildly non-IID scenario and 98.36% under strongly non-IID conditions, highlighting the effectiveness and reliability of the proposed framework for secure IoT-based DDoS attack detection.
Authors - P.N. Deorukhakar, V.B. Waghmare, I.K. Mujawar, R.Y. Patil Abstract - Convolutional Neural Networks (CNNs) have been widely and successfully applied to bioacoustic and passive acoustic monitoring tasks, including soundscape classification. However, the high dimension ality of CNN-derived embeddings often results in increased computa tional cost and reduced efficiency, particularly in iterative learning frame works such as Active Learning (AL) and in scenarios with limited labeled data. This work addresses these limitations by proposing a method for adapting CNN architectures to generate compact and discriminative em beddings tailored to soundscape data classification. The proposed ap proach leverages transfer learning and incorporates three progressively reduced dense layers (512, 256, and 128 neurons), enabling dimensional ity reduction to be learned intrinsically during network training rather than applied as a post-processing step. Experimental evaluations con ducted across multiple soundscapes datasets under the Active Learning paradigm demonstrate that the proposed embeddings consistently out perform conventional CNN embeddings (CNNE) in terms of classification performance and the efficient use of labeled data. These results indicate that integrating dimensionality reduction directly into CNN training en hances representation quality and robustness, offering an effective solu tion for soundscape data classification in labeling-constrained environ ments.
Authors - Domenico D’Uva Abstract - Indoor air quality (IAQ) is a frequently overlooked determinant of health in rural villages, where the extensive use of solid fuels for cooking and space-heating generates elevated concentrations of airborne pollutants. This study presents an integrated, low-cost protocol for improving IAQ in rural dwellings, combining real-time environmental monitoring, simplified digital modelling and passive strategies of ventilation and biophilic design. The methodology can be structured into three steps: Conceptual digital twin, feedback interface, ventilation strategies, biophilic integration. Conceptual digital twin is based on the mapping of each dwelling linked to Arduino low-cost, stand-alone sensors (CO₂, PM₂.₅, temperature and relative humidity) that collect data at temporal resolution of one minute. An immediate feedback interface based on visual and/or acoustic indicators that prompt residents to take corrective actions (selective opening of windows, activation of cross-breezes), when exposure thresholds - derived from WHO Air Quality Guidelines - are exceeded. Data-driven natural-ventilation strategies – optimal ventilation windows identified through time-series analysis of sensor data, calibrated to local weather conditions and occupancy profiles to maximise air exchange while minimising heat losses. Biophilic integration implies the introduction of resilient plant species with proven phytoremediation capacity, as Epipremnum aureum) which could reduce CO₂ level, with quantitative guidance on density (two to three plants per main room) and optimal placement. Using low-cost IoT sensors, the protocol monitors environmental parameters and pollutant concentrations in real time. The system targets specific safety and comfort thresholds, aiming to maintain CO₂ levels below 700 ppm and PM₂.₅ below 50 μg/m³ to optimize occupant health (Wu et al, 2021). These thresholds, derived from World Health Organization (WHO) guidelines, are essential to ensure occupant satisfaction and well-being. The ultimate objective is to define a scalable and replicable intervention model capable of combining digital technologies and natural solutions for the sustainable regeneration of fragile territories.
Authors - Kritika Singhal, Khushi Madeshiya, Utkarsh Upadhyay, Siser Pratap Singh, Surendra Kr. Keshari, Veepin Kumar Abstract - The integration of artificial intelligence in the academic en vironment has been rapidly growing since late 2022. One of the most widely adopted artificial intelligence tools in engineering is the large lan guage model. By using large language models, the engineering students can generate assignment answers, solve problems through code, and ex plain engineering concepts. Unlike traditional approaches, the large lan guage models can reduce time and simplify the students’ work. Many researchers have worked on artificial intelligence tools, most specifically large language models for engineers. This paper reviews the literature on the application of artificial intelligence tools in the following five areas of engineering education, which include programming, problem-solving in the core subjects, intelligent tutoring, technical writing, and simula tion support. Further, this paper discusses the main challenges of large language models in engineering education. Finally, this article concludes by outlining the future scope of large language models in engineering.
Authors - S.Venkata Rakesh, K.Tarun Kumar, A.Lohith, M.Nirupama Bhatt Abstract - One of the world's most destructive types of malware is ransomware, which results in huge financial and data loss around the globe. Current signature-based detection methodologies do not work for the detection of these types of ransomware because they have no way to identify them prior to their creation (zero-day) or when a variant of the ransomware is created (polymorphic). A behaviour-based ransomware detection methodology that involves the use of CPU Hardware Performance Counters (HPC) in combination with machine learning models for the purpose of detecting ransomware activity is the focus of this project. The following HPC metrics will be used to monitor the execution of a program or application while it is executing: instruction count; cache references; cache hits; branch instructions; and CPU cycles. These low-level architectural events will provide information on the unique behaviour characteristics of a ransomware program or application based on the types of behaviours exhibited by the encryption pro-cesses of a ransomware program or application. A labelled dataset of HPC traces of typical programs/applications will be developed by running both standard pro-grams/applications and ransomware in a controlled testing environment. Several supervised learning models such as Random Forest, Support Vector Machines, and Logistic Regression will be trained and validated on the labelled dataset. The experimental results show that ransomware activity causes significantly different HPC metrics, thereby allowing the correct identification of ransomware. The pro-posed methodology will offer a real-time, graphical user interface for real-time monitoring and graphical representation of the detected ransomware program or application.
Authors - Vasavi Ravuri, S. Lalitha Geetanjali, T. Bhavana Sri, V. Praveen, M. Mokshgna Teja Abstract - Unstructured vehicle traffic (i.e. those containing multiple users such as automobile drivers, pedestrians, cyclists, and even animals) creates a significant challenge for road safety. This work presents the development of a real-time road risk assessment (RRA) system for analyzing dashcam video that combines several computer vision techniques: object detection, semantic segmentation, multi-object tracking, and alert classification, into a unified, integrated processing pipeline. Object detection and multi-object tracking are accomplished using the YOLOv8m and ByteTrack with Kalman Filter algorithms. Additionally, semantic segmentation of the road scene is achieved using a SegFormer-B2. Finally, a segmentation-assisted fusion filter and perspective-aware danger zone are applied (to define each point in the field of view as belonging to a zone with certain levels of risk). The Road Intrusion Risk Score (RIRS) is a composite score that quantifies the severity of intrusion accumulated over time, and provides graduated alert levels. Testing of the system on COCO val2017 and four dashcam videos produced reliable object detections with significantly fewer false positives and very close to real-time performance, demonstrating the potential of the system to improve driver assistance systems in unstructured road environments.
Authors - Nathula Dayarathne, Guhanathan Poravi Abstract - This paper presents a novel methodology for predicting bug severity and priority in software development using machine learning models. The approach involves leveraging a manually curated dataset labelled with the support of industry experts, enabling the incorporation of domainspecific knowledge into feature selection and classification. A K-Means clustering method is initially employed to label the collected data, ensuring accurate grouping and feature extraction. The study identifies and utilizes 16 key features for classification and develops separate models for severity and priority prediction. These models, trained on the expertly labelled dataset, achieve high performance with accuracy metrics above 90%. This study uniquely combines K-Means pre-labelling with expert validation to reduce manual annotation while maintaining model accuracy. The proposed method demonstrates the effectiveness of combining clustering techniques with expert-driven labelling for improving bug management processes. By automating severity and priority classification, this research contributes to enhancing the efficiency and reliability of software development workflows.
