R Santhana Krishnan
@scadengineering.ac.in
Assistant Professor
SCAD College of Engineering and Technology
RESEARCH, TEACHING, or OTHER INTERESTS
Artificial Intelligence, Engineering, Computer Science, Agricultural and Biological Sciences
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
Scopus Publications
R. Santhana Krishnan, S. Gopikumar, A. Essaki Muthu, J. Relin Francis Raj, D. Abitha Kumari, and P. Stella Rose Malar
IEEE
In various regions, the maintenance of manholes is imperative, given the potential consequences for public health and safety. Neglecting this responsibility can result in severe outcomes, including loss of lives and the spread of diseases within the community. To address these challenges, an advanced anomaly detection system is proposed, leveraging deep learning with Autoencoders. Notably, the entire system operates on solar power, aligning with a commitment to environmentally sustainable practices. At the core of the system’s functionality is the Autoencoder, a deep learning model to differentiate complex manhole conditions. The Autoencoder is trained on normative sensor data, capturing nuanced patterns inherent in routine manhole states. This training enables the Autoencoder to perform real-time anomaly detection by swiftly identifying deviations such as the presence of noxious gases, irregular sewage levels, or unauthorized attempts at access. When anomalies are detected, signaling potential hazards, the Autoencoder triggers alert messages disseminated to municipal authorities via a cloud server. This proactive approach empowers timely intervention, particularly critical in scenarios involving toxic gases, ensuring the safeguarding of maintenance personnel. Additionally, the system integrates an NFC reader, streamlining authorized access for designated personnel while upholding robust security measures. In essence, the incorporation of Autoencoders into the system furnishes an intelligent anomaly detection infrastructure for manhole surveillance. By continually adapting to normative patterns, the Autoencoder enhances the system’s capacity to identify and promptly communicate potential risks, thereby improving the overall safety and efficiency of manhole maintenance practices.
R. Niranjana, T. Hemadarshana, S. Ilakkya, R.Santhana Krishnan, J.Jeniksha Epziba, and T. Preetha
IEEE
Skin disorders are a major worldwide health concern that require sophisticated diagnostic technologies for prompt and accurate identification. A deep learning approach that utilizes the DenseNet architecture for dermal image-based skin disease prediction is proposed. Creating and organizing a heterogeneous dataset covering a variety of skin conditions is part of the technique. Testing, verification, and training sets of the dataset are separated out to provide a thorough assessment of the model’s efficacy. Following its initial training on a large-scale dataset using pre-learned weights, the selected DenseNet model is refined using the skin disease dataset. Transferring expertise allows models to better identify relevant characteristics by utilizing information acquired from general photo recognition tasks. To adapt the architecture for the classification of skin diseases into many classes, custom layers are added. During the training phase, the model is optimized using suitable learning rates and loss functions. To reduce overfitting and improve the model’s generalization, hyperparameter adjustment is done. To assess the model’s performance, parameters like precision and loss are applied to the verification dataset. The model is evaluated on an alternative dataset to see if it can generalize to cases that haven’t been discovered yet once it has been successfully trained and validated. Throughout the process, patient privacy and data security are two of the most important ethical factors. The developed skin disease prediction model has the potential to be used in practical settings, facilitating prompt and precise diagnosis. Subsequent research endeavors could encompass ongoing enhancements to the model, cooperation with healthcare experts for clinical verification, and incorporation into user-friendly programs to ensure broad accessibility within healthcare environments.
P. Ebby Darney, N. Vallileka, Sumitha Manoj, A. Vegi Fernando, R. Santhana Krishnan, and S. Ram Prasath
IEEE
In contemporary vehicular communication systems, Vehicular Ad Hoc Networks (VANETs) serve a crucial role in enabling seamless data exchange among vehicles and infrastructure components. However, the dynamic and unpredictable nature of vehicular environments renders VANETs susceptible to various faults and disruptions, potentially compromising network performance and jeopardizing safety-critical applications. Therefore, the development of robust fault prediction and recovery mechanisms is imperative to ensure the reliability and resilience of VANETs. This research introduces a novel Hybrid Fault Prediction and Recovery Framework for VANETs utilizing the paradigm of Artificial Intelligence (AI) and Federated Internet of Things (IoT). The proposed algorithm, Deep Neural Network with Federated IoT Model (DNN-FIOT), is presented to enhance fault prediction and recovery efficacy in VANET environments. Extensive simulation analyses comparing DNN-FIOT with existing algorithms are conducted, employing suitable simulation metrics. Results demonstrate the superior performance of DNN-FIOT in terms of fault prediction accuracy, recovery time, and network stability. The proposed framework offers a promising solution for robust fault management in VANETs, ensuring uninterrupted vehicular communication and safety.
S. Jegadeesan, R.K Hayvita, A.G Nishath, S. Ramalakshmi, R. Santhana Krishnan, and A. Essaki Muthu
IEEE
Data synchronization is a critical aspect in the deployment of Digital Twins within Fog-Edge Cloud environments, ensuring consistency and reliability across distributed systems. However, the security and efficiency of synchronization techniques remain significant challenges. The paper presents a novel algorithm named TwinCrypt, which is designed based on the combination of appropriate encryption algorithm called as Homomorphic Encryption and suitable computing mechanism scheme called as Secure Multi-Party Computation. The algorithm is aimed to achieve secure and efficient data synchronization for Digital Twins in Fog-Edge Cloud environments. The proposed algorithm aims to address vulnerabilities associated with data transmission and storage, offering robust protection against unauthorized access and data breaches. The performance effectiveness of TwinCrypt is analysed by conducting suitable simulations and the simulation results are compared with are conducted comparing it with the existing algorithms used for data synchronization viz. RSA Encryption, AES Encryption, and Zero-Knowledge Proofs in similar environments. Simulation metrics such as latency, throughput, and security overhead are employed to assess the performance of TwinCrypt against its counterparts. The results establish that TwinCrypt outperforms existing algorithms in both security and efficiency, showcasing its suitability for real- world deployment scenarios.
J. Relin Francis Raj, J. Ebens Nikshya, S. Vinothini, R Umesh, R. Santhana Krishnan, and S. Gopikumar
IEEE
In contemporary urban landscapes, the quest for sustainable energy consumption has propelled the evolution of smart building technologies. These structures are equipped with advanced sensors, actuators, and control systems to optimize energy utilization while ensuring occupant comfort and operational efficiency. Central to this paradigm is energy modeling, a process that entails the analysis and prediction of energy consumption patterns within building environments. Efficient energy modeling holds paramount significance for devising effective strategies to mitigate energy wastage, reduce operational costs, and minimize environmental footprints. This research presents a novel approach, the Hierarchical Deep Learning Framework for Multi-scale Energy Modeling in Smart Building Environments, aimed at enhancing energy modeling efficiency. The Deep Energy Fusion (DEF) algorithm is introduced by integrating Deep Belief Networks (DBNs) with a hierarchical energy modeling architecture. DEF leverages DBNs' capability to learn complex hierarchical representations of energy data across various scales, facilitating a comprehensive understanding of energy consumption patterns. To evaluate DEF's performance, the simulation analysis are conducted by comparing it with existing algorithms. Metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R^2) are employed to assess prediction accuracy. Our results demonstrate that DEF outperforms traditional algorithms in capturing both local and global dependencies within the energy data, thereby improving energy modeling accuracy. Moreover, DEF showcases superior performance in terms of computational efficiency and scalability. This research contributes to advancing energy modeling techniques in smart buildings, offering a robust framework for optimizing energy management strategies
P. Kalyanakumar, S. Srinivasa Pandian, S. Boopalan, D. Kani Jesintha, R. Santhana Krishnan, and A. Essaki Muthu
IEEE
This research investigates the integration of bio-inspired optimization and swarm intelligence principles with TinyML for the development of energy-aware Internet of Things (IoT) devices. A novel model algorithm, termed "BioSwarmML," is introduced and evaluated against existing algorithms through comprehensive simulation analyses employing suitable metrics. The proposed framework aims to enhance energy efficiency in IoT applications by leveraging the collective intelligence derived from bio-inspired optimization and swarm behaviors. The "BioSwarmML" algorithm is designed to draw inspiration from natural processes, incorporating bio-inspired optimization techniques such as genetic algorithms, simulated annealing, and evolutionary strategies. Concurrently, swarm intelligence principles are integrated to emulate decentralized and self-organizing behaviors observed in biological systems. This amalgamation aims to optimize the energy consumption of TinyML models on IoT devices, facilitating sustainable and adaptive learning processes. Simulation analyses involve a comparative study with established algorithms, evaluating BioSwarmML based on metrics such as energy consumption, accuracy, and latency. The results demonstrate the efficacy of the proposed algorithm in achieving superior energy efficiency while maintaining competitive performance in terms of accuracy and responsiveness. The comparison sheds light on the advantages of BioSwarmML in energy-aware TinyML applications, showcasing its potential for widespread adoption in IoT ecosystems. This research contributes to the advancement of energy-efficient IoT systems by introducing a novel algorithmic paradigm that aligns with international journal standards. The proposed "BioSwarmML" model showcases a promising avenue for enhancing sustainability in TinyML-driven IoT applications, offering a valuable addition to the existing body of knowledge in the field.
M. Subramanian, K. Gokulakrishnan, B. Vijaya Nirmala, S. Vijay Shankar, R. Santhana Krishnan, and G. Vinoth Rajkumar
IEEE
The proposed patient monitoring system is a substantial progression in healthcare technology. It integrates various sensors and communication modules to provide real-time patient data to caregivers and medical professionals. The array of sensors includes a non-contact temperature sensor, a rain sensor, a Max30100 sensor, and a load cell sensor. The non-contact temperature sensor vigilantly tracks the patient's body temperature, alerting caregivers to any deviations. The rain sensor serves as a communication channel for patients unable to express needs, detecting urination. The Max30100 sensor monitors critical health metrics like heart rate and blood oxygen levels for early detection of conditions. The load cell sensor observes saline levels in IV bags, triggering alerts on nearing critical thresholds. Powered by an Arduino microcontroller, the system aggregates data, relaying it to a dedicated server via Wi-Fi for remote accessibility. The Isolation Forest-based IsoSentry Algorithm enhances anomaly detection, identifying aberrations in health readings. Adding reliability, a GSM module connected to Arduino dispatches real-time alerts via text messages. In essence, "GuardianForest" epitomizes a transformative leap in healthcare, integrating cutting-edge hardware, wireless communication, and cloud-based data management to enhance overall healthcare responsiveness and efficacy.
S. Jegadeesan, S. S. Aishwarya, R. G. Harshini, A. Justin Diraviam, S. Sujatha, and R. Santhana Krishnan
IEEE
The ever-expanding Internet of Things (IoT) landscape poses significant challenges for network reliability and resilience. This research investigates the design and implementation of a resilient Software-Defined Networking (SDN) framework with Internet of Things (IoT) integration, aiming to enhance network reliability. The proposed model, referred to as HybridSDN-IoT Resilience (HyRes), introduces a hybrid SDN approach by amalgamating traditional and distributed SDN strategies. This hybridization seeks to balance the advantages of centralized control and distributed resilience within the network architecture. To estimate the efficacy of the HyRes model implementation, comprehensive simulation analyses are conducted, comparing its performance against established algorithms in the SDN domain. Metrics such as network latency, packet loss, and throughput are employed to assess the reliability of the network under varying conditions. The simulation results offer significant performance results for the resilience and adaptability of HyRes in comparison to existing algorithms, showcasing its potential to mitigate network failures and enhance overall reliability. The proposed HyRes model not only establishes a foundation for resilient SDN architectures but also discourses the precise challenges in the integration of diverse IoT devices. By emphasizing adaptability and responsiveness, the research advances the discourse on the practical realization of reliable SDN frameworks in the context of emerging IoT technologies.
R. Santhana Krishnan, S. Balamurugan, M. Rekha, R Arul Jose, K. Haribabu, and A. Essaki Muthu
IEEE
In response to the increasing environmental threats, ForeScanGuard emerges as a vital forest monitoring system. Integrating advanced sensors and a proactive anomaly detection algorithm, it employs a 5-channel fire sensor for early fire detection, ultrasonic and vibration sensors for identifying tree cutting activities, and a BMP120 sensor to confirm tree falls. GPS modules attached to trees allow real-time location tracking, transmitted via Wi-Fi to a cloud server. The ForeScanGuard Algorithm analyzes sensor data for anomalies like fire outbreaks, illegal logging, and tree falls. Its proactive alerts facilitate rapid responses from forest officials, ensuring effective forest management. ForeScanGuard, with multiple sensors and cloud-based analytics, represents a significant advancement in smart forest monitoring, promoting conservation efforts through real-time anomaly detection and informed decision-making. This comprehensive solution enhances ecosystem protection, fostering a sustainable and proactive approach to forest conservation.
R. Santhana Krishnan, J. Relin Francis Raj, V. Bibin Christopher, N. Nandini, A. Essaki Muthu, and C. Puvanadevi
IEEE
This research explores the augmentation of Agricultural Internet of Things (IoT) systems through the integration of advanced predictive analytics and reinforcement learning models. A novel algorithm, termed "CropQL," is proposed to optimize crop yield prediction and resource allocation in precision agriculture. CropQL amalgamates the strengths of the CropNet IoT model with Q-Learning for reinforcement learning and Random Forest for predictive analytics. Simulation analyses are conducted to evaluate the performance of CropQL against established algorithms, employing diverse metrics such as precision, recall, F1 score, and area under the receiver operating characteristic curve (AUC-ROC). The proposed CropQL algorithm exhibits superior predictive accuracy, demonstrating a significant enhancement in crop yield forecasting when compared to baseline algorithms. Precision and recall metrics reveal the algorithm's efficacy in minimizing false positives and negatives, ensuring precision agriculture systems make informed decisions. F1 score indicates the balanced optimization of precision and recall, reinforcing the algorithm's robustness. AUC-ROC analysis further corroborates the superior discriminative power of CropQL in distinguishing between positive and negative instances. In comparison to existing algorithms, including AgriSensNet, PrecisionCropNet, and IoTFarmGuard, CropQL consistently outperforms in predictive accuracy and resource optimization. The simulation results underscore the efficacy of the proposed algorithm in enhancing the efficiency of agricultural IoT systems, providing a promising avenue for sustainable precision farming.
I. Sakthidevi, S. Vijaya Shankar, R. Santhana Krishnan, Sincy Elezebeth Kuruvilla, M. Bharath, and S. Gopikumar
IEEE
This research study introduces an IoT-Enabled Cradle System with the "DreamFlowRNN" algorithm, combining Recurrent Neural Networks (RNN) and Flow-based models. Equipped with smart sensors, the cradle monitors infants' physiological and environmental data during sleep. DreamFlowRNN analyzes this information, classify sleep stages, detect disturbances, and predict future patterns. The proposed system aims to provide predictive parenting capabilities by offering valuable insights into infants' sleep behavior. Extensive simulations, comparing DreamFlowRNN to existing algorithms like LSTM, CNN, and SVM, reveal its superior performance in accurate sleep stage classification, disturbance detection, and generation of realistic synthetic sleep patterns. The proposed system stands out with higher F1-score, precision, inception score, and lower Frechet Inception Distance (FID) value, emphasizing its efficacy in enhancing the well-being of infants through advanced AI-driven sleep pattern analysis.
R. Sheela Daniel, S. Sundararajan, C. Antony Vasantha Kumar, J. Nirmal Jothi, R. Santhana Krishnan, and C. Ramasamy Sankar Ram
IEEE
India is often heralded as the "Cynosure" of the global dairy industry, boasting a staggering valuation of INR 14,899.8 billion in 2022. This industry has exhibited remarkable and sustained growth in milk product availability since the year 1998. However, it's imperative to concede that the dairy sector concurrently bears the dubious distinction of being one of the most ecologically burdensome industries, primarily due to its substantial effluent output. Approximately 50% of the total dairy wastewater emanates from the rinsing of processing units intricately involved in the production of dairy commodities. The sources of this wastewater are multifarious, encompassing spills and seepage from pipelines, suboptimal tanker drainage, foaming incidents, tank spillages, cleaning operations, and liquid losses. This wastewater comprises milk solids, detergents, sanitizers, milk remnants, and cleaning water. It is typified by elevated concentrations of nutrients, both organic and inorganic, and may also encompass acids, alkalis, active agents, disinfectants, in conjunction with a substantial microbiological load, inclusive of pathogenic viruses and bacteria. In recognition of the imperative need to reclaim valuable milk solids to reinvigorate agriculture, particularly in regions grappling with water scarcity, a comprehensive wastewater treatment strategy was proffered. This strategic framework encompassed the meticulous scrutiny of pivotal parameters such as pH, BOD (Biological Oxygen Demand), and turbidity during the treatment regimen. Moreover, an ongoing discourse pertains to the utilization of natural coagulants for the efficacious treatment of dairy wastewater. Diverse ratios of these coagulants were judiciously introduced to the wastewater under treatment, and the levels of pH, BOD, and turbidity were assiduously assessed upon the culmination of each processing stage. The determination of the treatment process's effectiveness hinged upon the discerning comparison of outcomes arising from the utilization of various reagents or natural coagulants in the treatment protocol.
M. Manicka Prabha, S. Jegadeesan, S. D. Jayavathi, G. Vinoth Rajkumar, J. Nirmal Jothi, and R. Santhana Krishnan
IEEE
This research study explores the transformative potential of IoT-Enabled Smart Mirrors to redefine home connectivity by enabling seamless internet browsing and smart home integration. It introduces a novel model algorithm, IoT-ViSense (IVS), which leverages IoT sensor data fusion with vision-based Convolutional Neural Networks (CNN) for real-time environmental monitoring and security. A comprehensive simulation analysis of IoT-ViSense is conducted and compared against existing algorithms using pertinent simulation metrics. IoT-Enabled Smart Mirrors serve as multifunctional interfaces, integrating personalized weather updates, news, calendar schedules, and real-time compliments. They also incorporate features such as a built-in camera for capturing high-quality selfies, on-demand workout sessions facilitated by fitness experts, and the ability to browse the internet, make calls, and interact with other connected devices. The proposed IoT-ViSense (IVS) algorithm combines IoT sensor data with Vision-based CNN analysis to enhance environmental monitoring and security. It continuously processes data from IoT sensors and camera feeds, detecting anomalies and patterns in real-time. By fusing these data streams, IVS enhances security by identifying unauthorized access and suspicious activities while concurrently optimizing resource usage and ensuring comfortable living and working environments. The simulation analysis of IoT-ViSense (IVS) is compared with suitable existing algorithms, assessing its efficiency in terms of accuracy, response time, and resource optimization. Results reveal that IoT-ViSense (IVS) outperforms other algorithms, demonstrating its prowess in achieving a harmonious blend of smart home integration and real-time environmental monitoring.
S. Sundararajan, P. Ebby Darney, K. Palanivel Rajan, A. Vegi Fernando, J. Nirmal Jothi, and R. Santhana Krishnan
IEEE
Securing valuable assets has become imperative in contemporary contexts, with a particular emphasis on safeguarding precious items such as jewellery. This research introduces a comprehensive solution for enhancing the security of jewellery boxes through a novel TripleFuse Recognition Algorithm (TFRA). TFRA integrates Convolutional Neural Networks (CNN) for Fingerprint Recognition, Support Vector Machines (SVM) for Barcode Recognition, and Recurrent Neural Networks (RNN) for OTP Authentication. The algorithm is coupled with an Internet of Things (IoT) model utilizing the Message Queuing Telemetry Transport (MQTT) protocol, facilitating efficient communication and data exchange between the jewellery box and the authentication components. To evaluate the effectiveness of TFRA, simulation analyses are conducted and compared with established algorithms in the field. The simulation metrics include accuracy, precision, recall, and F1-score. TFRA exhibits superior performance in achieving high accuracy rates, ensuring robust Fingerprint Recognition, Barcode Recognition, and OTP Authentication. Furthermore, the proposed model demonstrates enhanced efficiency in communication and data transfer through the MQTT protocol within the IoT framework. Comparative analyses with existing algorithms underscore TFRA's efficacy, particularly in achieving a harmonious synergy between diverse recognition tasks. The results indicate that TFRA not only surpasses its counterparts in terms of authentication accuracy but also offers an optimized solution for real-time location tracking within the secured jewellery box. This research contributes to the burgeoning field of IoT-based security applications, providing a comprehensive solution for safeguarding valuable assets. The proposed TFRA algorithm, integrated into an advanced IoT model, demonstrates promising results in the context of three-way authentication and location tracking for secured jewellery boxes.
Ram, L. Narayanan and S. Krishnan
University of the Aegean
<p>Trees in the forest are an unprecedented cluster of organisms in ecological, monetary and social importance. With a wide distribution, mostly random spread over and a large population in terms of size, the majority of tree species show considerable variation in genetics both within and between populations. The genus Phytophthora, is a most destructive plant pathogen and attacks a wide range of tree hosts, including inexpensively its significant species. Many species of Phytophthora are known to be persistent and brought in through nurseries and commercial agriculture. Diseases from various eucalyptus tree species were first reported in India and the symptoms, incidence and leaf damage have been described. As an observation of a larger project to use genomic data for tree disease diagnosis, pathogen detection and surveillance, in this study the significant analysis of various DNA sequences of P.meadii 2 on eucalyptus tree species in Indian Forests. The identified outcomes observed that the high frequency of nucleotides and their combinations found in the organisms in trees threatening their lives may be observed and has to be condensed</p>
P. Sundara Vadivel, B. Karthika, Y. Harold Robinson, R. Santhana Krishnan, L. Rachel, and S. Sundararajan
IEEE
In the retail commercial sector, various advancements in IoT and AI have industrially renovated the domain. Due to the introduction of boosted connectivity and availability of intelligent systems pave way for redefined shopping experience. The current research work highlights a pioneering tactic to improve shopping experiences. It integrates an intelligent IoT-driven Perceptive shopping cart and reinforcement learning techniques for optimized store navigation. In the research work, the salient features of the two existing deep learning algorithms viz. Reinforcement Learning and Q Network with Long Short-Term Memory are integrated and an innovative algorithm called Reinforced Q-Network with Long Short-Term Memory (ReQL-Net) is proposed. The ReQL-Net algorithm authorizes the perceptive cart to unconventionally acquire effectual navigation strategies. Extensive simulation results show that The ReQL-Net algorithm outperforms other reinforcement learning or Q-Network methods in terms of Navigation Efficiency, Exploitation Trade-off and Convergence Rate. Using the proposed algorithm, the customer can personalize his shopping experience. Also the shopkeeper can designed for individual customer experiences. This information can be used to create targeted marketing strategies. Also, such researches on smart shopping cart opens up new possibilities for the future of intelligent shopping systems which can adapt to changing retail environments.
I. Sakthidevi, A. Srinivasan, C. Gayathri, E. Golden Julie, R. Santhana Krishnan, and P. Ebby Darney
IEEE
In today’s world, cardiovascular diseases including heart attacks are the prime reason for mortality. Detecting heart attack risk at early stages and accurate assessment of such risks related to cardiovascular conditions play an essential role in active patient management and preventive interventions. In the current research work, a novel deep learning framework called "CardioSightFrame" is proposed for detecting heart attack risks and cardiovascular conditions at early stages using retinal images. The proposed algorithm integrates Graph Convolutional Networks (GCNs) and Vision Transformers (ViTs) to control both local structural information and global contextual understanding from the retinal scans. A comprehensive simulation analysis is conducted to evaluate the performance of CardioSightFrame and it is compared with the existing algorithms. Also, the predictive accuracy of the proposed algorithm is measured using appropriate simulation metrics. By treating retinal images as graph nodes and leveraging self-attention mechanisms, the proposed algorithm achieves accurate and efficient feature extraction which is vital for early detection of heart attack risk and cardiovascular conditions. To evaluate CardioSightFrame’s performance, wide-ranging simulation experiments are conducted using a varied dataset of retinal images. Through these comparisons, the predictive accuracy, sensitivity, specificity, and computational efficiency of CardioSightFrame are established. The simulation results indicate that CardioSightFrame outclasses existing algorithms in terms of accuracy and sensitivity for early detection of heart attack risk and cardiovascular conditions. The algorithm’s ability to efficiently investigate retinal images with reduced computational overhead further enhances its applied pertinence.
J. Nirmal Jothi, N Soundiraraj, P. Ebby Darney, R. Santhana Krishnan, K. Lakshmi Narayanan, and S. Sundararajan
IEEE
Movie review sentiment analysis has gained significant attention due to the vast amount of user-generated content available on social media platforms like Twitter. Accurately analyzing sentiment from such data can provide valuable insights into public opinion and assist in decision-making processes. In the current work, a novel machine learning model, called Sent2Bayes, for movie review sentiment analysis on Twitter is proposed. The proposed model combines the power of Word2Vec embeddings and the Multinomial Naive Bayes classifier to enhance sentiment prediction accuracy. The Sent2Bayes model begins with the creation of Word2Vec embeddings, which capture semantic relationships between words and encode contextual information. These embeddings are then fed into the Multinomial Naive Bayes classifier, which leverages probabilistic modelling to predict sentiment labels for movie reviews. By combining the strength of Word2Vec's semantic understanding and Multi-nomial Naive Bayes' probabilistic approach, Sent2Bayes aims to improve the accuracy and robustness of sentiment analysis on Twitter data. To evaluate the performance of Sent2Bayes, this study conducts an extensive simulation experiments comparing it with existing algorithms commonly used for sentiment analysis. Further, this study employs suitable simulation metrics, including accuracy, precision, recall, and confusion matrix, to measure the model's predictive capabilities and generalization ability. The simulation results demonstrate that Sent2Bayes achieves superior sentiment analysis performance compared to SVM and RNN. The proposed model shows higher accuracy, precision, recall, and confusion matrix across various datasets and test scenarios. The improved performance of Sent2Bayes effectively captures contextual information through Word2Vec embeddings and the robust probabilistic modelling of Multinomial Naive Bayes.
N. Vallileka, P. Sundaravadivel, U. Karthikeyan, R. Santhana Krishnan, K. Lakshmi Narayanan, and S. Sundararajan
IEEE
With the increase in the popularity of the social media sites, there arises lot of adverse effects including the spread of fake news. This research study proposes a novel algorithm, called DeepTweet, which leverages transformer-based models for enhanced fake news detection in Twitter sentiment analysis. The aim is to improve the accuracy and efficiency of identifying fake news in real-time by exploiting the contextual information and semantic representations captured by the transformer-based models. The proposed DeepTweet algorithm utilizes a pre-trained transformer model, such as BERT or GPT, to encode the textual content of tweets and extract rich contextual embeddings. These embeddings capture both local and global dependencies, enabling a more comprehensive understanding of the sentiment expressed in a tweet. Furthermore, a novel attention mechanism tailored for fake news detection is introduced to focus on key textual features that are indicative of potential misinformation. To assess the efficiency of DeepTweet, extensive simulations using a large dataset of labelled tweets containing both genuine and fake news are done. Further, this study compares the performance of DeepTweet with existing algorithms, including LSTM, CNN, and traditional machine learning approaches. Various simulation metric measures such as accuracy, precision, recall, and F1 score are employed to analyze the algorithm's performance. The simulation grades demonstrate that DeepTweet outperforms the prevailing algorithms by means of accuracy and overall detection performance. The proposed attention mechanism proves to be particularly effective in identifying subtle linguistic cues and deceptive patterns associated with fake news. The enhanced contextual embeddings captured by the transformer-based models contribute to improved sentiment analysis and subsequently lead to more accurate fake news detection.
I. Sakthidevi, G. Vinoth Rajkumar, R. Sunitha, A. Sangeetha, R. Santhana Krishnan, and S. Sundararajan
IEEE
The rapid advancement of machine learning (ML) and artificial intelligence (AI) has created an increasing demand for efficient and automated processes in training and deploying AI models. In cloud environments, where vast computational resources are available, orchestrating the entire lifecycle of machine learning workflows becomes crucial to leverage the scalability and flexibility offered by the cloud infrastructure. This research study proposes a novel system architecture and simulation model for machine learning orchestration in cloud environments, aiming to automate the training and deployment of using Distributed Machine Learning (DML) AI model. The proposed system architecture consists of three key components: Job Manager, Resource Manager, and Model Repository. The Job Manager handles the scheduling and coordination of machine learning tasks, ensuring efficient resource allocation and utilization. The Resource Manager dynamically manages the allocation and provisioning of computing resources based on workload demands. The Model Repository acts as a centralized repository for storing and versioning AI models, enabling seamless model deployment and updates. To evaluate the effectiveness and performance of the proposed system architecture, a simulation model is developed. The simulation model provides a virtual environment that mimics real-world cloud scenarios, allowing for extensive experimentation and analysis. Various performance metrics such as training time, resource utilization, and scalability are measured and compared against baseline approaches to demonstrate the superiority of the proposed system architecture. The simulation results indicate that the machine learning orchestration system in the cloud environment significantly improves the efficiency and automation of training and deploying Distributed Machine Learning (DML) AI model. The proposed architecture optimizes resource allocation, minimizes training time, and enhances scalability, leading to cost savings and increased productivity. Moreover, the simulation model provides valuable insights into the behaviour and performance of the system under different workload scenarios, facilitating the fine-tuning and optimization of the orchestration process.
C. Ashokkumar, M. Arun Kumar, R. Santhana Krishnan, S. Mohanap Priya, K. Lakshmi Narayanan, and E. Golden Julie
IEEE
Animal intrusion is a significant challenge for farmers, causing extensive crop damage, human injuries, and substantial financial losses. Traditional animal movement monitoring and surveillance methods alone are insufficient to provide a permanent solution. To solve this problem, a novel system architecture that combines the latest convolutional neural network (CNN) algorithm and Internet of Things (IoT) technology is proposed. Our proposed system architecture integrates various components, including a Raspberry Pi as a central processing unit, cloud storage for efficient data management, and a GSM module for instant alert generation. To train the CNN algorithm, a comprehensive and diverse Animal Dataset consisting of various animal species commonly found in farmland areas is curated. The dataset encompasses a wide range of annotated images, enabling the CNN algorithm to accurately identify and classify animals. The Raspberry Pi serves as the core of the system, responsible for real-time image processing and analysis. Utilizing the power of the CNN algorithm, the Raspberry Pi processes the captured images from strategically placed surveillance cameras. When an animal intrusion is detected, the system promptly generates an alert via the integrated GSM module, providing immediate notifications to farmers and relevant authorities. Furthermore, the system leverages cloud storage to store and manage the collected data, facilitating easy access and retrieval for analysis and system improvement. This cloud-based approach enables scalability, allowing the system to handle large amounts of data efficiently. By integrating the CNN algorithm, IoT, Raspberry Pi, cloud storage, and GSM module, a comprehensive and robust framework for real-time detection and prevention of animal intrusion is provided. The system's ability to swiftly identify and alert farmers and authorities about potential threats minimizes crop damage, ensures human safety, and significantly reduces financial losses.
K. Rajkumar, A. Valli, R. Santhana Krishnan, G. Karpagarajesh, S. Sundararajan, and K. Lakshmi Narayanan
IEEE
To combat the environmental impact of charcoal usage in iron carts, the Smart Laundry Cart System (SLCS) is presented. SLCS utilizes a solar-powered smart iron box, eliminating the need for charcoal and mitigating tree cutting. The system integrates a user-friendly mobile application for effective customer management, enabling iron cart owners to streamline their operations efficiently. In the event of solar power unavailability, SLCS seamlessly switches to the main power supply, ensuring uninterrupted ironing services. Additionally, the system facilitates online payments, promoting digital transactions and enabling secure record-keeping for future reference. SLCS revolutionizes iron cart operations, reducing reliance on charcoal and advancing sustainability in the ironing process. By adopting SLCS, the estimated 10 million individuals involved in this occupation can contribute to preserving natural resources and embracing a more eco-friendly approach.
S. Nithya, X. Arogya Presskila, B. Sakthivel, R. Santhana Krishnan, K. Lakshmi Narayanan, and S. Sundararajan
IEEE
Sentiment analysis, the intricate task of discerning and classifying the myriad of sentiments conveyed within textual data, has captured substantial interest and intrigue, primarily driven by the pervasive utilization and influence of social media platforms. In this study, a novel approach to enhance sentiment analysis of Twitter data by employing Recurrent Neural Networks (RNNs) with an attention mechanism is proposed. The proposed model leverages the sequential nature of tweets and the attention mechanism to capture the inherent dependencies between words and highlight salient information. The RNN-based model on a large-scale dataset of annotated Twitter data, encompassing diverse sentiments is trained. The model effectively learns the contextual information and sentiment patterns, enabling accurate sentiment classification. A comprehensive set of tests were run to evaluate the effectiveness of this methodology, and the outcomes were meticulously compared to those of traditional machine learning algorithms and established deep learning models. The empirical findings demonstrate that proposed attention-based RNN model performs better than competing methods and achieves cutting-edge performance on sentiment analysis of Twitter data. Moreover, an in-depth analysis of the attention weights generated by the model, shedding light on the significant words and phrases influencing sentiment classification is conducted. This provides valuable insights into the underlying sentiment dynamics in Twitter data. Proposed research contributes to the field of sentiment analysis by proposing an effective and robust approach for Twitter sentiment classification. The findings highlight the potential of RNNs with attention mechanisms in capturing the nuanced sentiment expressions prevalent in social media text. The proposed model can facilitate various applications, including real-time sentiment monitoring, brand reputation analysis, and public opinion tracking, benefiting industries and researchers alike.
A. Essaki Muthu, A. Justin Diraviam, R. Niranjana, K. Saravanan, S. Sundararajan, and R. Santhana Krishnan
IEEE
Modern banking relies heavily on Automated Teller Machines (ATMs), although card-based transactions have historically faced security issues. This study proposes Near-Field Communication (NFC) Card Emulation and fingerprint technology as secure alternatives for ATM cash transactions. NFC Card Emulation enables secure data exchanges within a short distance, ideal for handling sensitive information. To augment security during authentication, fingerprint technology is proposed as an additional layer of protection. By employing fingerprint sensors, user identification can be established without reliance on physical cards or NFC tags. This integration ensures a heightened level of security during the authentication process, surpassing the current card-based approach employed in ATMs. The system eliminates the need for physical cards, enhancing security and mitigating risks such as ATM tampering and magnetic strip damage. A high level of security is guaranteed during ATM transactions as authentication becomes more effective and secure. This novel strategy strengthens ATM security and provides users with a seamless and safe banking experience by incorporating NFC Card Emulation and fingerprint technologies.
J. Allwyn Kingsly Gladston, P. Kalyanakumar, K. Rajkumar, R. Santhana Krishnan, R. Niranjana, and S. Sundararajan
IEEE
Food security, poverty reduction, and rural livelihood support are all greatly impacted by agriculture. To fight crop damage and water scarcity, a smart agricultural system built on the Internet of Things has been created. The system keeps an eye on the water and soil moisture levels and guards against unauthorized animal entry. Cell phones can be used by users to get real-time data. The water pump operates automatically on and off depending on criteria measured on the farmland. For the benefit of farmers and rural communities, the solution attempts to maximize irrigation, protect crops, and increase agricultural output. This smart farming system uses technology to give farmers and rural communities a dependable and effective tool that has a favorable effect on agricultural activity. A revolutionary solution that reduces risks and boosts productivity in the agricultural industry is offered through the use of IoT technology. The approach greatly enhances food security, promotes sustainable development in rural regions, and benefits farmers' well-being.