Personalised Emotion-Based Email Composer Using Large Language Models B. Aarthi, B.V. Haswath, Arjun V. Nair, S. Buvaneshwar, Pankajdeer Bikumalla, A. Prabha, Szeberényi András Multidisciplinary Advancements in Human AI Augmentation, 2025 Reading and expressing emotional context in written text is vital for successful interaction in digital communication. This project presents a Personalised Emotion-Based Email Composer that produces emotionally sensitive and grammatically correct emails through Natural Language Processing (NLP) and deep learning approaches. The system utilizes transformer-based models, such as BERT for emotion classification, T5 for email generation and summarization, and GPT-based models for contextual content generation. Emotion recognition is achieved through fine-tuned models based on benchmark datasets such as Go Emotions and ISEAR, which provide rich annotations of emotional status across various textual inputs. The system parses user input to determine the emotional tone, producing a grammatically correct and empathetic email draft following the deduced or declared sentiment. There is a final polishing step to confirm grammatical correctness and contextual relevance. The system has good utility in personal and business communication by automating the production of emotionally intelligent responses. This research bridges the gap between affective computing and automated content generation, contributing to the development of emotionally intelligent AI systems.
Combating Deepfake-generated photos and videos using generative adversarial network B. Aarthi, A. Smruthi, Pamireddy Thanishka, G. Sakthi Prasanna, P. Mahendran Pioneering AI and Data Technologies for Next Gen Security Iot and Smart Ecosystems, 2025 Rapid advances in artificial intelligence and machine learning have resulted in the creation of Deep Fakes, which are manipulated films, audio, and images capable of disseminating false information, fake news, and altering sensitive records. The prevalence of deepfake technology has raised significant concerns regarding the veracity of digital content, underscoring the critical need for reliable deepfake facial recognition algorithms. This study embarks on developing an advanced deepfake detection system leveraging Generative Adversarial Networks (GANs) within a programming environment. The central focus is to create a neural network that can effectively differentiate between authentic and artificially generated media content. To accomplish this, the system undergoes extensive training using diverse datasets, enabling it to recognize subtle nuances and specific artifacts associated with GAN-generated content.
IoT technologies 5G and 6G Charanjeet Singh, M. Prasad, B. Aarthi, Ranjeet Yadav, Nidhya M. S., Balajee Maram Revolutionizing Data Science and Analytics for Industry Transformation, 2025 Optical wireless technologies have garnered significant attention in recent years due to their potential to address the growing demands of 5G and 6G networks. These technologies promise to revolutionize physical-world applications by providing enhanced energy efficiency, higher speeds, and cost-effective solutions. However, their implementation faces numerous challenges, particularly in the context of sensors deployed in the internet of things (IoT). This chapter examines critical issues such as the limitations of wireless optical communication, frequent handovers, inter-cell interference losses caused by atmospheric variations, and the need to mitigate flickering effects. By analyzing these challenges, the authors aim to contribute to the development of robust and sustainable optical wireless communication systems, paving the way for their seamless integration into next-generation IoT ecosystems.
Multilingual language classification model for offensive comments categorisation in social media using HAMMC tree search with enhanced optimisation technique B. Aarthi, Balika J. Chelliah International Journal of Computational Science and Engineering, 2025 The exponential rise of social media platforms has led to a surge in offensive content, highlighting the necessity for effectively detecting and managing such comments. This necessitates precise and advanced online social networks (OSN) categorisation and optimisation methods. This study introduces and assesses a novel technique for automatically categorising texts, supporting over 60 languages, without relying on a pre-annotated dataset. The technique employs multilingual methods based on the randomised explicit semantic analysis (ESA) strategy. To combat the inherently multilingual nature of social media content, the paper introduces an innovative classification and optimisation strategy named 'hybrid adaptive Markov chain Monte Carlo tree search (HAMCMTS) with enhanced eagle Aquila optimiser (EEAO)'. The study uses three publicly available datasets to identify negative or offensive comments in various languages, offering a comprehensive analysis in this field. The proposed approach holds potential for diverse applications, particularly in multilingual categorisation tasks like monitoring disaster-related communications on social media to improve visibility and trust. Moreover, it incorporates a sophisticated mechanism to bolster the dependability of its recommendations.
AI-Powered Emergency Response System for High-Risk Zones Using Real-Time Video, Audio, and Geolocation Analysis Vinayak Pai, Kamaleshwaran S, Vinston Jose C, Aarthi B 2025 IEEE Pune Section International Conference Punecon 2025, 2025 Emergency response systems are needed in highrisk zones such as tension regions, attack-prone zones, or even refugee camps exposed to attacks against civilians or military personal, but they reveal a challenge of unpredictable threats and barriers to reporting and communication. This paper examines and compares four video classification methods for distress detection in a high-risk zone - VideoResNet, Hierarchical ResNetLSTM (Improved), R2P1D-HierNet, and MC3-HierNet. Models leverage ResNet50 or similar backbones for feature extraction, with variations in temporal and hierarchical processing. R2P1DHierNet achieves the highest weighted F1-score (0.77), while MC3-HierNet balances anomaly recall (0.837 binary). The process of building on these experiences after previous deployments in conflict zones ensures that intent detection is automated, manages the data, and decreases response times and workloads for responders. Moreover, the system provides context that can ensure adaptation to existing (old and new) infrastructures that can additionally increase wider adaptation to multiple high-risk zone environments.
Dynamic Traffic Optimization Through Cloud-Enabled Big Data Analytics and Machine Learning for Enhanced Urban Mobility M. Nithin Kishore, Aarthi B, R. Prabavathi, Rohith S., Bhavan Karthick R. P 2025 International Conference on Computing and Communication Technologies Iccct 2025, 2025 Due to the need to solve urban mobility problems caused by traffic congestion, traffic consumes much energy, brings environmental harm and economic losses. This research proposes an intelligently automated traffic control system enabled by cloud computing and big data analytics to further enhance urban mobility while maximizing traffic efficiency. The system obtains and processes real-time traffic data through GPS devices, sensors and social media through Random Forest algorithms applied to predictive modeling. The system provides dynamic traffic control strategies including adaptive signal timings and route optimization which are responsive to changes in real time. A cloud based infrastructure is used for handling your data and scaling efficiently. Real time inputs and monitoring and vehicle simulation is performed over a user friendly interface for predictive analysis and testing purposes. This integrated approach promises to lead to reduced congestion, decreased travel times, and enhanced traffic safety using an affordable, scalable, and flexible urban traffic management solution.
Next Gen Road Safety: Harnessing YOLOv11 Explainable AI and Cloud Aware Context Nivethitha E, Adithi V, T S Sowmya, Vinoth R, Flavia B, Aarthi B 2025 International Conference on Emerging Technologies in Engineering Applications Icetea 2025 Proceedings, 2025 Road safety is a critical concern, and unsurfaced roads are a special case where potholes and cracks are among the greatest hazards to human-driven and autonomous vehicles. To follow through with the issue of tackling road hazards in real time, we propose a real-time road hazard detection system that incorporates deep learning, explainable AI (XAI) techniques, and context-aware detection to enhance the Advanced Driver Assistance Systems (ADAS). The YOLOv11 model is used to detect road anomalies quickly and easily available for hazard identification. YOLOv11 was selected after comparative evaluation with YOLOv8, Single Shot Detector (SSD), and Attention based Convolutional Neural Networks due to its higher detection accuracy, faster inference and better performance under occlusion. It outperformed other models in terms of mean Average Precision (mAP), reduced false positives, and maintained superior inference speed under occlusion, shadow, and low-light conditions. Furthermore, its architecture supports efficient edge deployment, crucial for real-time ADAS systems. Edge detection, coupled with occlusion sensitivity analysis, yields context-aware detection, which detects partially exposed potholes and cracks, thereby reducing false negatives and providing the overall robustness of the system under difficult road conditions. Stereo vision with the use of multi- camera fusion methodology is used to estimate depths and effectively handle occlusions caused by other vehicles as well as provide better spatial localizations of hazards. The field of view of the vehicle is used to divide the location of the hazards into left, centre, and right zones. A real-time voice alert mechanism helps in acting ahead of time in a larger perspective of situational awareness. The proposed system is cloud-integrated in order to achieve large-scale deployment and to continuously keep improving the model with model updates and sharing of the hazard data in the cloud at remote nodes for collective intelligence. This research integrates deep learning, explainable AI, and contextual analysis, rooted in knowledge garnered from the intrinsic knowledge of its generation, and builds a future generation of knowledge-enabled intelligent transportation systems for the accurate, interpretable real-time road hazard detection for applications of ADAS and drivers at large, improving road safety for all users.
Analyzing the forest fire using correlation methods Chandrasegar Thirumalai, B Aarthi, V Abhinaya Proceedings of the International Conference on Electronics Communication and Aerospace Technology Iceca 2017, 2017
Agriculture facilitation using cloud computing - Survey International Journal of Pharmacy and Technology, 2016
