Sathya Subramanian

Sathya Subramanian

Auto-generated profile, based on publicly available information

Sathya Subramanian

S.A. Engineering College

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 September 2022 in Indonesian Journal of Electrical Engineering and Computer Science
Verified
Journal Article
Open access
PDF file
TATsehay Admassu
Tsehay Admassu
TKThulasi Karpagam
Thulasi Karpagam
SSSathya Subramanian
Sathya Subramanian
SJSenthil Murugan Janakiraman
Senthil Murugan Janakiraman
JAJayanthi Arumugam
Jayanthi Arumugam
DADawed Omer Ahmed
Dawed Omer Ahmed
The existing heart failure risk prediction models are developed based on machine learning predictors. The objective of this study is to identify the key risk factors that affect the survival time of heart patients and to develop a heart failure survival prediction model using the identified risk factors. A cox proportional hazard regression method is applied to generate the proposed heart failure survival model. We used the dataset from the University of California Irvine (UCI) clinical heart failure data repository. To develop the model we have used multiple risk factors such as age, anemia, creatinine phosphokinase, diabetes history, ejection fraction, presence of high blood pressure, platelet count, serum creatinine, sex, and smoking history. Among the risk factors, high blood pressure is identified as one of the novel risk factors for heart failure. We have validated the performance of the model via statistical and empirical validation. The experimental result shows that the proposed model achieved good discrimination and calibration ability with a C-index (receiver operating characteristic (ROC) of being 0.74 and a log-likelihood ratio of 81.95 using 11 degrees of freedom on the validation dataset.

6 December 2025 in International Journal of Environment, Engineering and Education
Verified
Journal Article
Open access
PDF file
KLK. Lokeshwaran
K. Lokeshwaran
KNKomal Kumar Napa
Komal Kumar Napa
JMJ. Senthil Murugan
J. Senthil Murugan
VEV. Elanangai
V. Elanangai
SSS. Sathya
S. Sathya
In today’s digital environment, online reviews have become one of the key factors that influence the decisions of customers. This is especially true in areas such as e-commerce, travel and the hospitality industry, where buyers depend heavily on the shared experiences of others before making a choice. At the same time, the growing issue of fake or fabricated reviews has raised serious concerns, as it reduces the reliability of online platforms and creates confusion for consumers. Detecting such misleading reviews is not an easy task, since the language used in them is often very close to what is seen in genuine opinions. In the present work, an attempt has been made to compare the performance of traditional machine learning techniques with that of transformer-based deep learning models for the identification of fake reviews. As part of the baseline, Logistic Regression and Linear SVM were applied with TF-IDF features. On the other hand, advanced architectures like BERT, RoBERTa and XLNet were fine-tuned on the Deceptive Opinion Spam Corpus. The results clearly indicated that the classical models gave accuracies in the range of mid-80 percent, whereas the transformer-based models performed much better, crossing or coming close to 90 percent. Among the transformer models, RoBERTa showed the most balanced performance across precision and recall, XLNet gave the highest recall, which is very important when sensitivity is the main concern, while BERT achieved competitive results with less demand on computing resources.

1 December 2025 in Bulletin of Electrical Engineering and Informatics
Verified
Journal Article
Open access
KNKomal Kumar Napa
Komal Kumar Napa
SMSangeetha Murugan
Sangeetha Murugan
SSSathya Subramanian
Sathya Subramanian
DSDurga Devi Saravanan
Durga Devi Saravanan
DNDevana Nageswari
Devana Nageswari
BPBattula Krishna Prasad
Battula Krishna Prasad
Accurate malaria detection is crucial for effective disease management, particularly in regions with limited medical resources. Deep learning models have shown promising results in automated diagnosis, yet real-world deployment often faces challenges such as computational cost and model interpretability. This study evaluates multiple deep learning architectures—VGG16, ResNet50, InceptionV3, MobileNetV2, and DenseNet121—on the publicly available National Institutes of Health (NIH) malaria cell image dataset (27,558 images), and enhances their performance using stacking ensemble learning with different meta-learners. Among individual models, DenseNet121 achieved the highest accuracy of 88.00%, while MobileNetV2 had the lowest at 84.80%. Implementing stacking with logistic regression as the meta-learner improved accuracy to 89.40%, while random forest increased it to 90.10%. The best performance was achieved with XGBoost as the meta-learner, attaining an accuracy of 91.20%, precision of 92.10%, recall of 90.80%, and an F1-score of 91.40%—representing a 3.2% accuracy improvement over the best individual model. The classification report further confirms superior performance in distinguishing infected and uninfected cases. These results highlight the potential of stacking with advanced meta-learners to support health workers in rapid, reliable malaria diagnosis, ultimately aiding timely treatment, and improving patient outcomes in clinical and field settings.

6 May 2026 in Discover Artificial Intelligence
Verified
Journal Article
Open access
PDF file
KNKomal Kumar Napa
Komal Kumar Napa
SMSangeetha Murugan
Sangeetha Murugan
SJSenthil Murugan J
Senthil Murugan J
SSS. Sathya
S. Sathya
ABA. G. Balamurugan
A. G. Balamurugan
NDNageswari D
Nageswari D
Coronary artery disease (CAD) is a leading cause of mortality worldwide, necessitating early, accurate, and non-invasive prediction methods. While machine learning offers a promising alternative to conventional invasive diagnostics, challenges such as high-dimensional data, class imbalance, and redundant features remain. This study proposes HESML-CAD, a hybrid ensemble stacking and meta-learning framework enhanced with Bald Eagle Search Optimization (BESO) and explainable AI for reliable and interpretable CAD prediction. The model was evaluated on two datasets: Framingham (4200 samples, 15 features) for 10-year risk prediction and Z-Alizadeh Sani (304 samples, 55 features) for current diagnosis. Preprocessing included imputation, normalization, encoding, and ADASYN-based balancing. Feature selection combined Chi-square, mutual information, autoencoder-based reduction, and BESO optimization. Five base classifiers (KNN, SVM-RBF, Random Forest, XGBoost, and CatBoost) were stacked using Logistic Regression, with Platt scaling for calibration and SHAP for explainability.On the Framingham dataset, HESML-CAD achieved an accuracy of 0.93, F1-score of 0.93, and AUC of 0.96, outperforming the best individual model (Random Forest: accuracy 0.896, F1-score 0.90) by approximately 3–4%, with a low Brier score of 0.0540. On the Z-Alizadeh Sani dataset, it achieved an accuracy of 0.94, F1-score of 0.935, and AUC of 0.98, surpassing strong baselines (accuracy ≈ 0.93), while maintaining well-calibrated probability estimates (Brier score 0.0547). SHAP identified clinically relevant predictors such as age, systolic blood pressure, and chest pain. Overall, HESML-CAD provides accurate, calibrated, and interpretable CAD predictions, demonstrating potential for supporting non-invasive clinical screening.

1 June 2026 in International Journal of Evaluation and Research in Education (IJERE)
Verified
Journal Article
Open access
KNKomal Kumar Napa
Komal Kumar Napa
RGRajkumar Govindarajan
Rajkumar Govindarajan
SSSathya Subramanian
Sathya Subramanian
SJSenthil Murugan Janakiraman
Senthil Murugan Janakiraman
NDNageswari Devana
Nageswari Devana
BMBilla Manindhar
Billa Manindhar
This study investigates the reading habits, genre preferences, and reading behaviors of undergraduate students and examines how these factors influence their reading performance. A total of 342 responses were directly collected from students through a structured questionnaire. Descriptive statistics revealed strong inclinations toward analytical genres such as mystery/thriller, science fiction, and true crime, while newspaper reading frequency remained low. Hypothesis testing showed no significant differences in reading scores across gender or academic departments. A significant positive correlation emerged between daily reading duration and newspaper reading frequency. Most notably, students who preferred analytical genres demonstrated significantly higher reading scores (Cohen’s d=1.36). Regression analysis further confirmed genre preference as the strongest predictor of reading performance. These findings highlight the importance of genre engagement and daily reading routines in enhancing reading comprehension and literacy development. The study offers meaningful implications for educators, curriculum designers, and reading intervention programs.

1 August 2026 in IAES International Journal of Artificial Intelligence (IJ-AI)
Verified
Journal Article
Open access
PDF file
LKLokeshwaran Kanagaraj
Lokeshwaran Kanagaraj
RPRaguraman Purushothaman
Raguraman Purushothaman
SSSathya Subramanian
Sathya Subramanian
DSDurga Devi Saravanan
Durga Devi Saravanan
KNKomal Kumar Napa
Komal Kumar Napa
BMBilla Manindhar
Billa Manindhar
Distributed denial of service (DDoS) attacks continues to pose serious risks to modern networks, with their growing intensity making early detection both critical and challenging. Conventional machine learning (ML) models often struggle with the nonlinear and highly dynamic nature of attack traffic, which motivates the use of advanced architectures. In this study investigate a transformer-based classifier for DDoS detection on the CIC-DDoS2019 dataset. The workflow included preprocessing, feature scaling, and domain-guided feature selection. Logistic regression (LR) was employed as a baseline, achieving 92.1% accuracy and F1-score of 0.90, thereby revealing the limitations of linear models. The transformer, after hyperparameter tuning and 5-fold cross-validation, reached an average accuracy of 99.95% with precision, recall, and F1-scores all above 99.9%. The model demonstrated stable convergence and generalization across folds. These results highlight the strength of attention mechanisms in capturing feature dependencies, while also pointing to future directions such as real-time deployment, explainability, and resilience to zero-day attacks.