A multi strategy optimization framework using AI digital twins for smart grid carbon emission reduction S. Sakthivel, M. Arivukarasi, G. Charulatha, J. Nithisha, B. Abirami, A. K. Jaithunbi, V. Suresh Kumar Scientific Reports, 2026 This research presents an AI-enabled digital twin framework to achieve carbon neutrality in smart grids through optimal management of heterogeneous energy storage systems. The proposed structure integrates battery, thermal, and hydrogen storage technologies with AI-driven forecasting models to address the challenge of renewable integration, while maintaining grid stability and economic viability. This paper presents a comparative analysis of three distinct optimization methodologies, like a rule-based (RB) heuristic approach, Model Predictive Control (MPC) with look-ahead capability, and a multi-objective Genetic Algorithm (GA). Simulation results that demonstrate the AI-optimized multi-energy storage (MES) integration significantly enhance the renewable utilization and reduce carbon emissions by approximately 30% compared to conventional approaches. Specifically, the MPC achieves a 29.9% reduction in carbon footprint (1741.1 kgCO₂ vs. 2485.2 kgCO₂ baseline) with corresponding operational cost savings of 30%, while GA shows a comparable 28.2% improvement. The comparative analysis discloses a critical trade-off between computational complexity, optimization performance, and practical implementability, with MPC emerging as a balanced method for a real-world application. This work has contributed to sustainable energy systems by providing a comprehensive framework for MES optimization, imparting treasured insights for grid operators and policymakers. The outcomes highlight the important role of AI-enabled digital twin in designing next-generation smart grid infrastructure, which is capable for supporting excessive renewable penetration at the same time as ensuring reliability and sustainable economic growth.
Federated autoencoder IDS for IoT: A Fed-ANIDS approach using CICIoT2023 M. Arivukarasi, S. Harihara Gopalan, S. Gnanamurugan, R. Dhanapal, A. Ramachandran, A. Manikandan Systems and Soft Computing, 2026 Intrusion detection on Internet of Things (IoT) networks is still a challenging task because of the vast quantity of data, evolving attack patterns, and strong privacy controls across the scattered systems. Even though the current deep learning and Transformer-based intrusion detection systems are effective, numerous IoT nodes cannot offer the centralized data aggregate or processing capabilities that these systems require. This paper introduces Fed-ANIDS, a federated autoencoder-based intrusion detection system enhanced with feature selection by using Learning-based Intelligent Intrusion Detection (LBIID) and a High-performance ViT Intrusion Detection System (HiViT-IDS) classifier. This model is based on a lightweight feature-selection module to reduce duplication before classification, autoencoder to generate small latent representations, and a federated training approach to safeguard data privacy. The complete analysis of the CICIoT2023 dataset allows concluding that the proposed solution has a high detection rate and a significantly reduced inference latency and overhead in communication. The results confirm the suitability of the proposed methodology to real-time, scalable and privacy protection IoT intrusion detection. The accuracy of the system was 99.84% with precision and recall at 98.86%, F1-score at 98.85% and specificity at 98.02%.
Hybrid AI-Driven Predictive Analytics Framework for Real-Time Patient Monitoring and Adaptive Early Warning Systems in Smart Hospitals Arivukarasi M Proceedings of 5th International Conference on Evolutionary Computing and Mobile Sustainable Networks Icecmsn 2025, 2025 Modern healthcare has been completely transformed by the quick development of artificial intelligence (AI) and the Internet of Medical Things (IoMT), which have made it possible for automated, intelligent, and continuous patient monitoring. This study suggests a Hybrid AI-Driven Predictive Analytics Framework that combines deep learning and machine learning models for adaptive early warning system (EWS) production and real-time patient monitoring in smart hospitals. IoMT-enabled wearables and bedside sensors are used to gather multimodal physiological data, including blood pressure, temperature, oxygen saturation, heart rate, and ECG signals. The hybrid AI approach uses Long Short-Term Memory (LSTM) networks for temporal sequence prediction and machine learning methods for trend-based anomaly detection to find early indicators of patient decline. Reducing false alarms and increasing diagnostic accuracy, an adaptive alert mechanism constantly modifies threshold values according to clinical risk categories and patient-specific health profiles. Computational efficiency is increased through cloud-edge integration, which guarantees scalability and low-latency analytics across hospital networks. According to experimental data, the framework outperforms conventional EWS techniques in terms of accuracy, precision, and sensitivity. By improving clinical decision-making, speeding up reaction times, and fortifying the digital infrastructure of smart hospitals, this research helps create proactive, intelligent, and patient centred healthcare systems.
A Methodical Approach to Creating Medical Imaging Datasets for Deep Learning-Powered Lung Disease Identification M Arivukarasi, J Nithisha Proceedings IEEE 10th International Conference on Smart Structures and Systems Icsss 2025, 2025 Recent developments in deep learning models have significantly accelerated the creation of automated decisionmaking systems, especially those designed to identify lung issues using medical imaging techniques. Large, varied, and highly labeled chest X-ray (CXR) and computed tomography (CT) databases would be necessary for the positive identification of respiratory diseases including COVID-19, pneumonia, and tuberculosis. This study provides an overview of the process of gathering data and creating datasets for deep learning-based lung disease detection systems. We perform efficient preprocessing and labeling operations, conduct a systematic analysis of various institutional and publically available data, and handle important concerns such class imbalance, inter-site variance, and ethical considerations. Our performance evaluation reveals that our developed data framework, which filters data, has high classification parameters with 0.985 AUC and 96.4 percent accuracy when it classifies into various classes of lung diseases. These results provide an excellent foundation for efficient classification systems and the clinical application of deep learning models to diagnostic systems.
A Fuzzy-based Neurodiagnostic System for Early Detection of Parkinson's Disease using Wearable Sensor Data Satyanarayana Nimmala, Ch Srilakshmi, Nukala Sujata Gupta, Pallati Narsimhulu, Saggurthi Kiran, M Arivukarasi Proceedings of 5th International Conference on Soft Computing for Security Applications Icscsa 2025, 2025 Parkinson's Disease (PD) is a neurodegenerative disorder that necessitates early diagnosis to facilitate effective intervention. This paper suggests a Fuzzy-Based Neurodiagnostic System that utilizes the UCI Parkinson's Telemonitoring Dataset to classify early PD cases by integrating fuzzy logic and a deep neural network. 22 biomedical voice features, such as average fundamental frequency, jitter, shimmer, noise-to-harmonics ratio (NHR), and non-linear dynamical measures such as DFA and PPE, are included in the dataset. These features are captured from sustained phonation recordings. The interpretability of vocal irregularities is improved by the fuzzy logic module, which assigns fuzzy membership values (low, medium, high) based on clinical thresholds while processing three important uncertainty-prone features—jitter, shimmer, and NHR. These fuzzified outputs, in conjunction with the normalized raw features, are fed into a four-layer fully connected neural network (FCNN) that utilizes ReLU and Tanh activations. The first hidden layer consists of 64 neurons, while the subsequent layers contain 32 and 16 neurons, respectively. For binary classification (PD or non-PD), the output layer implements a Sigmoid activation. The model is trained using binary cross-entropy loss and 10-fold cross-validation, as well as the Adam optimiser. It outperformed conventional classifiers, including SVM and Decision Trees, with 94.8% accuracy, 92.3% precision, and 96.1% recall in this study. This hybrid soft computing approach effectively models uncertainty in voice features and extracts non-linear patterns, making it a promising diagnostic tool for real-time, noninvasive Parkinson's screening using ubiquitous sensor data.
Embedded System for Real-Time Weather Monitoring and Accurate Rainfall Prediction using Sensors Prabakaran S, Sabi L, Bharathi P, Arthi K, Pavithra R, Arivukarasi M Proceedings of 3rd International Conference on Intelligent Cyber Physical Systems and Internet of Things Icoici 2025, 2025 Weather monitoring and forecasting is both essential in agriculture, disaster management and dealing with climate adaptation. But traditional systems usually have disadvantages of being very expensive, not readily available and devoid of localized real-time forecast, hindering their competence in meeting the menace of global warming. To address these drawbacks, the present system suggest a weather station system based on the IoT technology that will combine soil moisture and DHT11 sensors with ESP32 microcontroller and Wi-Fi module. The data captured is sent to the ThingsSpeak IoT analytic platform, which could be monitored through the ThingsView Android application remotely. Other than real-time monitoring, a deep learning framework of Gated Recurrent Unit (GRU) is used to forecast rainfall based on sensor data with better results compared to conventional techniques. The opportunities to plan and prepare agricultural land and mitigate disasters are facilitated by the system due to the provision of continuous, low-cost, and easy access of environmental information. The findings of this work prove a scalable IoT-AI solution to localized weather forecasting presenting direct societal and environmental value.
Hybrid Bee-Ant Colony Model for Real-Time Network Resource Allocation in 6G Environments Satyanarayana Nimmala, K Purushotham, T Suvarna Kumari, Guguloth Ravi, M Arivukarasi, Nithisha J Proceedings of 8th International Conference on Computing Methodologies and Communication Iccmc 2025, 2025 This paper proposes a Hybrid Bee-Ant Colony Optimization (HBACO) method to efficiently and quickly assign network resources in 6G environments. The method suggested uses both the global search and solution exploration of the ABC algorithm, as well as the ACO algorithm's local refinement and path optimization. The framework consists of two cycles: (i) the Bee Phase, where scout and onlooker bees design and test different solutions for bandwidth, power and computational resource allocation and (ii) the Ant Phase, where ant agents use pheromones to refine the solutions and find the best routing and allocation paths under shifting network demands. The simulations were done in MATLAB R2023b using the ITU AI/ML in 5G Challenge’s 6G-IoT dataset, including mobility, latency, and real-life traffic information. The 6G mmWave environment consists of 100 mobile users and 20 small cells, all operating at 28 GHz using a 500 MHz bandwidth. The evaluation shows that HBACO achieves an 18.7% increase in throughput, a 26.4% decrease in latency, and a 21.2% better energy efficiency than baseline ACO, ABC, and Genetic Algorithms. The HBACO model, with a fairness index of 0.93, is a flexible, intelligent, and scalable way to allocate resources in ultra-dense 6G networks.
Efficient Phishing Detection and Prevention Using Support Vector Machine (SVM) Algorithm M Arivukarasi, A Manju, R Kaladevi, Shanmugasundaram Hariharan, M Mahasree, Andraju Bhanu Prasad Proceedings 2023 12th IEEE International Conference on Communication Systems and Network Technologies Csnt 2023, 2023 Phishing issues influence the electronic trade in light of the fact that web-based clients trust the Internet climate less. Phishers use procedures that advance to bait online clients, making new phishing sites and spreading messages that attempt to persuade Internet clients to follow deceitful connections to get to their sites. Phishing sites utilize refined procedures that direct internet-based clients to open another page, which has not yet been added to the boycott. A phishing assault that utilizes these new sorts of strategies is known as a zero-day assault. Against phishing techniques can be isolated into specialized or non-specialized arrangements. Nontechnical arrangements used to safeguard the client from phishing assault rely upon utilizing mindfulness and preparing projects to show online customers how to perceive phishing messages and sites. Specialized arrangements, in any case, rely upon building recognition and security models in view of preparing datasets.
A comprehensive survey of deceitful conclusion and counteractive action in multimodal datasets utilizing data mining and machine learning Journal of Advanced Research in Dynamical and Control Systems, 2019
Advanced template deduction in heterogeneous web pages for price comparison International Journal of Applied Engineering Research, 2014
Distributed dos attack in IP spoofing using symmetric block cipher technique International Journal of Applied Engineering Research, 2014
Ws-business transactions by using rule based technique in distributed environment International Journal of Applied Engineering Research, 2014
A new model for biometric highly secure authentication using distributed mobile system International Journal of Applied Engineering Research, 2014
