Article Per Year (5 Year)
p-Index From 2021 - 2026
0.751
P-Index
This Author published in this journals
All Journal Emerging Science Journal
Basant Kumar
Department of Mathematics and Computer Science, Modern College of Business and Science, Bowshar, Muscat
Environmental Science

Published : 4 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 4 Documents
Search

 1 
Evaluating Differential Privacy Mechanisms in Machine Learning with Emphasis on Utility and Robustness Rashmi Dwivedi; Basant Kumar; Vivek Mishra; Hothefa Jassim; Ozlem Kilickaya
Emerging Science Journal Vol. 10 No. 2 (2026): April
Publisher : Ital Publication

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.28991/ESJ-2026-010-02-07

Abstract

Federated learning enables collaborative model training across distributed clients without sharing raw data, yet it remains susceptible to inference threats such as membership inference attacks. This study aims to enhance the privacy of federated learning by integrating differential privacy and systematically evaluating its effects on model utility and adversarial robustness. A synthetic multimodal dataset was developed by combining demographic attributes from the UCI Adult dataset, mobility indicators from Google COVID-19 Mobility Reports, and semantic descriptors from LAION-400M, creating a high-dimensional and bias-reduced benchmark for privacy-preserving experimentation. Differentially private stochastic gradient descent (DP-SGD) was applied under multiple privacy budgets and ablation settings to isolate the individual contributions of gradient clipping and noise injection. Experimental results reveal that model accuracy increases with larger privacy budgets, while membership inference attack accuracy remains close to random guessing, confirming strong defense capability. Gradient clipping proved essential for training stability, whereas excessive noise caused measurable degradation in learning utility. The proposed framework establishes reproducible benchmarks for tuning differential privacy parameters in federated environments and demonstrates that robust privacy guarantees can be achieved without substantial loss of performance, providing practical guidance for deploying trustworthy, privacy-preserving machine learning systems across domains such as healthcare, finance, and mobility.
ZigBee Based Low Latency IoT and AI Integrated Framework for Real Time Telehealth Monitoring Basant Kumar; Mohammad Shahnawaz Shaikh
Emerging Science Journal Vol. 10 No. 2 (2026): April
Publisher : Ital Publication

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.28991/ESJ-2026-010-02-024

Abstract

The Internet of Things (IoT) and Artificial Intelligence (AI) have opened up new frontiers in remote health monitoring with the integration of technologies and transformative solutions in order to detect real-time health monitoring and disparities. This article shows an innovative and integrated wireless health surveillance system, which is aimed at auxiliary environments, especially for elderly and chronically ill patients. The system links IoT sensors to monitor heart rate, body temperature, and oxygen level with cloud-based AI-driven systems for continuous real-time health monitoring of data shared by IoT sensors. Taking advantage of the ZigBee protocol for low-power, reliable communication ensures spontaneous data transmission from a system wearer to a centralized processing unit. At the most basic level, the system uses advanced machine learning algorithms such as random forest, support vector machine (SVM), and logistic regression to identify health discrepancies with a high degree of accuracy. The random forest model in particular gets an impressive 95% accuracy and recalls 100%, ensuring reliable detection of minimum false negatives and important health issues. The modular structure of the system allows for the addition of more sensors, including blood pressure and glucose monitors, to ensure scalability and adaptability to suit the varying needs of different patients. In a real-world care facility, strict testing was carried out on the capability of monitoring the capacity system with just a 120 ms delay and a power consumption of 3.8 mW/h, which made it very suitable for long-term, energy-skilled deployment. By addressing some of the major issues such as high delays, false alarms, and lack of integration in current systems, this research provides a scalable, reliable, and user-friendly solution for telehealth. The proposed system not only adds more accuracy and freedom to the patient in the clinical setting but also lessens the burden of the healthcare providers, paving the way to a new generation of intelligent health solutions.
Leveraging Hybrid Deep Q-Learning for Early Identification of At-Risk Students P. Vijaya; Joseph Mani; Basant Kumar; Hothefa Shaker; Rajeev Rajendran
Emerging Science Journal Vol. 9 (2025): Special Issue "Emerging Trends, Challenges, and Innovative Practices in Education"
Publisher : Ital Publication

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.28991/ESJ-2025-SIED1-06

Abstract

Student performance prediction is employed to predict the learning performance to identify at-risk students. However, prediction models should also consider external factors along with learning activities, such as course duration. The student’s performance gets affected, which leads to a high decreasing rate and meets the risk of failing to complete the course on time. To overcome these challenges, this work proposed a Sea Lion Search Optimization (SLnSO) based on the Deep Q network (DQN) for predicting at-risk students. Here, the input data is taken from the dataset and forwarded to the data transformation phase, which is performed by Yeo-Johnson (YJ) transformation. Then, in the feature selection stage, the most relevant features are selected using the Damerau-Levenshtein technique. Then, Data Augmentation (DA) is performed to increase the dimension of the features, which is followed by the Deep Q Network (DQN) that is utilized for predicting the students at risk. Finally, by implementing the proposed SLnSO, the predicted results will be executed by DQN. The SLnSO-DQN is the combination of both Sea Lion Optimization (SLnO) and Squirrel Search Algorithm (SSA). The outcomes of the proposed model SLnSO-DQN attain significant performance that is based on various parameters, such as Mean Absolute Error (MAE), Mean Square Error (MSE), and Root MSE (RMSE), and also obtained better values of 0.327, 0.265, and 0.514, respectively.
Fractional White Smell Agent Optimization for CNN-Based Transfer Learning in Melanoma Classification Vijaya P; Basant Kumar; Joseph Mani; Satish Chander; Roshan Fernandes; Mohamed Sirajudeen Yoosuf
Emerging Science Journal Vol. 10 No. 3 (2026): June
Publisher : Ital Publication

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.28991/ESJ-2026-010-03-02

Abstract

Melanoma is the deadliest form of skin cancer, and early diagnosis and treatment can significantly reduce mortality rates. However, existing strategies for classifying melanoma from dermoscopic skin images still face significant challenges. Therefore, this study aims to develop an accurate method for melanoma classification using dermoscopic skin images. A novel melanoma classification framework, termed Fractional White Smell Agent Optimization-enabled Convolutional Neural Network-based Transfer Learning (FWSAO_CNN-based TL), is proposed. First, the input skin image is preprocessed using an Adaptive Kalman Filter. Subsequently, skin lesion segmentation is performed using LinkNet, where the network is trained using White Smell Agent Optimization (WSAO). Following segmentation, image augmentation is applied, and feature extraction is conducted. Finally, melanoma classification is performed using a CNN-based transfer learning model trained with the proposed Fractional White Smell Agent Optimization (FWSAO), which integrates the Fractional concept, Smell Agent Optimization (SAO), and White Shark Optimizer (WSO). The CNN utilizes hyperparameters derived from a pretrained GoogLeNet model. The performance of the proposed FWSAO_CNN-based TL framework was evaluated using accuracy, True Positive Rate (TPR), and True Negative Rate (TNR). The proposed method achieved values of 91.565%, 90.090%, and 91.269%, respectively. Furthermore, the proposed model demonstrated performance improvements of 18.4%, 8.1%, 17.5%, 12.55%, 8.2%, and 6.23% compared with conventional approaches.
Co-Authors Hothefa Jassim Hothefa Shaker Joseph Mani Mohamed Sirajudeen Yoosuf Mohammad Shahnawaz Shaikh Ozlem Kilickaya P. Vijaya Rajeev Rajendran Rashmi Dwivedi Roshan Fernandes Satish Chander Vijaya P Vivek Mishra