Dalvin Vinoth Kumar
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Dalvin Vinoth Kumar has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- Ashok M. Kumar
Computer Science and Software Engineering, Skyline University Nigeria, Kano City, Kano State, Nigeria
1 shared publication - Jeevanand E. S.
Christ University
1 shared publication - K. R. Rakshita
Christ University
1 shared publication - M. Senthil Vadivu
Christ University
1 shared publication - Pal Sanjoy Kumar
Department of Biological Sciences SSIT, Skyline University Nigeria Kano City, Kano State, Nigeria
1 shared publication Priyank Goyal
1 shared publication
Latest Publications
- 2 September 2026
- Published by Springer Nature
- p. 12-21
Inflation Rate Forecasting of India Using a Hybrid ARIMA-Bidirectional LSTM with Attention Framework and XGBoost Residual Correction
- 2 September 2026
- Published by Springer Nature
- p. 12-21
Inflation Rate Forecasting of India Using a Hybrid ARIMA-Bidirectional LSTM with Attention Framework and XGBoost Residual Correction
- Open Access
- 30 June 2026
- Published by Elsevier in SSRN Electronic Journal
Fine-grained textile provenance recognition: Deep learning and graph-based directions forIndian saree classification
Classifying Indian sarees in terms of their provenance can be framed as a fine-grained pattern recognition problem in which different provenances may have similar motifs, pallus, borders and layout combinations and within a provenance, there is a wide variation in the types of saree elements. This paper presents a concise review and a research agenda for classifying Indian sarees by provenance. It looks into the relevant works on convolutional and transformer-based learning, fine-grained recognition, metric and angular margin learning and graph neural networks. Existing textile and saree-related studies address issues such as textile pattern classification, fabric recognition and authentication. However, the studies reviewed do not explicitly address provenance recognition as a task from a multi-cluster perspective. This review identifies five gaps: limited dataset scale, reliance on imageonlymodels, weak modelling of relational textile knowledge, limited interpretability and insufficient multimodal integration. This review argues that a saree provenance classification system may benefit from combining visual embeddings with structured textilemetadata and graph neural networks.
- 30 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessFine-grained textile provenance recognition: Deep learning and graph-based directions forIndian saree classification
Classifying Indian sarees in terms of their provenance can be framed as a fine-grained pattern recognition problem in which different provenances may have similar motifs, pallus, borders and layout combinations and within a provenance, there is a wide variation in the types of saree elements. This paper presents a concise review and a research agenda for classifying Indian sarees by provenance. It looks into the relevant works on convolutional and transformer-based learning, fine-grained recognition, metric and angular margin learning and graph neural networks. Existing textile and saree-related studies address issues such as textile pattern classification, fabric recognition and authentication. However, the studies reviewed do not explicitly address provenance recognition as a task from a multi-cluster perspective. This review identifies five gaps: limited dataset scale, reliance on imageonlymodels, weak modelling of relational textile knowledge, limited interpretability and insufficient multimodal integration. This review argues that a saree provenance classification system may benefit from combining visual embeddings with structured textilemetadata and graph neural networks.
- Open Access
- 1 January 2025
- Published by National Library of Serbia in Yugoslav Journal of Operations Research
- Vol. 35 (4) , 895-918
Prediction of type 2 diabetes using support vector machine with enhanced levy flight based Fruitfly optimization algorithm and feature selection approaches
Researchers have been leveraging various data analytics methods for Diabetes mellitus (DM) diagnosis, prognosis and management. The data analytics paradigm has become advanced and automated with the emergence of machine learning (ML) and deep learning (DL) algorithms. With new techniques, the prediction accuracy of ML models for various real-world problems has increased significantly. In our previous work, we introduced and investigated the Improved K-Means with Adaptive Divergence Weight Binary Bat Algorithm to create an innovative diagnosis system. Across several problem scenarios, the performance of this algorithm is much better in terms of speed. However, this algorithm's accuracy of data categorization comes below expectations. To achieve high classification accuracy, the objective of this study work is to concentrate on methods and strategies. This aim is fulfilled through a Support Vector Machine (SVM) with an Enhanced Levy Flight-based Fruitfly Optimization model. This novel model improves diabetes prediction accuracy and can be applied to regressions, classifications, and other tasks. The nearest training data points? distances should be greater as this can lower classifiers? generalization errors. Missing values in datasets are retrieved using the Adaptive Neuro Fuzzy Inference System (ANFIS). A new algorithm called the Enhanced Inertia Weight Binary Bat Algorithm (EIWBBA) is introduced to optimize feature spaces and eliminate unimportant aspects. Further on, a novel feature selection technique is introduced by using the Enhanced Generalized Lambda Distribution Independent Component Analysis (EGLD-ICA). The classification uses a Support Vector Machine with an Enhanced Levy flight-based Fruitfly Optimization Algorithm (SVM-ELFFOA). The SVM-ELFFOA classification techniques are implemented using MATLAB software. It is evident that the discussed IKM-EIWBBA+SVM-ELFFOA classifier produces much better values of the accuracy of 93.50%, while the available IKM-EIWBBA+SVM yields 91.87%, IKM-ADWFA+LR renders 90.50%, and IKM+LR renders just 85.00%. From the simulation experiment, the proposed classification techniques implemented in MATLAB software and according to comparative data, this suggested model has a higher prediction accuracy of 93.50% compared to existing classification methods.
- 1 January 2025
- Published by National Library of Serbia in Yugoslav Journal of Operations Research
- Vol. 35 (4) , 895-918
Open AccessPrediction of type 2 diabetes using support vector machine with enhanced levy flight based Fruitfly optimization algorithm and feature selection approaches
Researchers have been leveraging various data analytics methods for Diabetes mellitus (DM) diagnosis, prognosis and management. The data analytics paradigm has become advanced and automated with the emergence of machine learning (ML) and deep learning (DL) algorithms. With new techniques, the prediction accuracy of ML models for various real-world problems has increased significantly. In our previous work, we introduced and investigated the Improved K-Means with Adaptive Divergence Weight Binary Bat Algorithm to create an innovative diagnosis system. Across several problem scenarios, the performance of this algorithm is much better in terms of speed. However, this algorithm's accuracy of data categorization comes below expectations. To achieve high classification accuracy, the objective of this study work is to concentrate on methods and strategies. This aim is fulfilled through a Support Vector Machine (SVM) with an Enhanced Levy Flight-based Fruitfly Optimization model. This novel model improves diabetes prediction accuracy and can be applied to regressions, classifications, and other tasks. The nearest training data points? distances should be greater as this can lower classifiers? generalization errors. Missing values in datasets are retrieved using the Adaptive Neuro Fuzzy Inference System (ANFIS). A new algorithm called the Enhanced Inertia Weight Binary Bat Algorithm (EIWBBA) is introduced to optimize feature spaces and eliminate unimportant aspects. Further on, a novel feature selection technique is introduced by using the Enhanced Generalized Lambda Distribution Independent Component Analysis (EGLD-ICA). The classification uses a Support Vector Machine with an Enhanced Levy flight-based Fruitfly Optimization Algorithm (SVM-ELFFOA). The SVM-ELFFOA classification techniques are implemented using MATLAB software. It is evident that the discussed IKM-EIWBBA+SVM-ELFFOA classifier produces much better values of the accuracy of 93.50%, while the available IKM-EIWBBA+SVM yields 91.87%, IKM-ADWFA+LR renders 90.50%, and IKM+LR renders just 85.00%. From the simulation experiment, the proposed classification techniques implemented in MATLAB software and according to comparative data, this suggested model has a higher prediction accuracy of 93.50% compared to existing classification methods.
- 3 January 2024
- Published by Springer Nature in Neural Computing and Applications
- Vol. 36 (10) , 5399-5412
CNN-based Indian medicinal leaf type identification and medical use recommendation
- 3 January 2024
- Published by Springer Nature in Neural Computing and Applications
- Vol. 36 (10) , 5399-5412
CNN-based Indian medicinal leaf type identification and medical use recommendation