MULTI-ASPECT SENTIMENT ANALYSIS IN AMAZON REVIEWS USING GMM-ENHANCED TEXTCAPS WITH PROBABILISTIC CAPSULE ROUTING Journal of Theoretical and Applied Information Technology, 2026
A Stacking Ensemble Federated Deep Learning Model with Optimization for the Efficient Ocular Pathology Detection S Geethamani, L Sankari, A Elsawy, M Abdel-Mottaleb, M Aboushousha, et al. Texila International Journal of Public Health, 2025 One major challenge in healthcare is utilizing Fundus Images (FI) to diagnose ocular pathology (OP).An ocular disorder disrupts the eye's regular functioning or adversely impacts the eye's visual acuity.Almost everyone experiences eye-sight issues throughout their lives, ranging from minor problems that can be managed at home to more severe conditions requiring specialized medical care.While certain kids require specialized care, others are minors who do not show up to support requests or who can be handled at home with ease.Ocular pathology detection approaches depend on Stacking Ensemble Federated (DL)Deep Learning (SEFDL), which was suggested in this work.First, an adaptive weight (AW)-based median filter (MF) is applied to image resizing and removing noise.Then, the data augmentation, coupled with the Synthetic Minority Over-sampling Technique (SMOTE), z, is employed to address data imbalance, a common issue in medical datasets.Finally, SEFDL is proposed for disease detection (DD).Adaptive TSO (Tuna-Swarm Optimization) Technique adjusted hyperparameters (HP) for 4 pre-trained models: CNN, VGG16, Inceptionv2, and ResNet50.DL models trained centrally have been compared with the enhanced algorithms in a federated framework.The proposed SEFDL model demonstrates superior accuracy and robustness when benchmarked against existing methods, highlighting its potential as a reliable diagnostic tool.Finally, the result must be compared with existing approaches to improve ocular pathology detection while addressing data privacy concerns in healthcare applications.
Stacked Denoising Autoencoder (SDAE) based Image Denoising and Ensemble Deep Learning Model (EDLM) for Ocular Pathology Detection L. Sankari, S. Geethamani Proceedings of the 7th International Conference on Innovative Data Communication Technologies and Application Icidca 2025, 2025 Automatic screening of fundus tissue images for visual abnormalities is a practical way to increase screening and solve the scarcity of skilled experts (such as retina specialists and ophthalmic surgeons) in light of aging and expanding populations. In recent years, it has advanced significantly, particularly as deep learning (DL) has advanced. In digital image processing, noise filtering is an essential step that serves as preprocessing. Thus, several images reconstruction difficulties might be solved using an efficient image denoising technique. In this paper, Stacked Denoising Autoencoder (SDAE) model is introduced for image denoising. SDAE learns robust, high-level features from noisy data by stacking multiple denoising autoencoders (DAEs) in layers. Enhanced Residual Network (ERNet) is used to provide precise characteristics and density in the network are extracted from ocular pathology images. Ensemble learning technique is introduced which combines the predictions from multiple models (Convolutional Neural Network (CNN), Visual Geometry Group-19 (VGG-19), Inception V4, and MobileNetV2). The method that organisms naturally perceive visual information served as the model for CNN. It has three completely connected and sixteen convolutional layers 19 weight layers of the deep convolutional neural network VGG-19. With Inception-ResNet, the Inception V4 network was revealed. MobileNetV2 is introduced inverted residual blocks and linear bottlenecks to its architecture, improving accuracy and speed. The suggested algorithm's effectiveness was assessed using the f-measure, accuracy, precision, and recall existing techniques. To enhance the diagnosis of ocular pathologies while resolving data privacy issues in healthcare applications, the result must eventually compare with present techniques.
Air Pollution Prediction by Deep Learning Model S. Jeya, L. Sankari Proceedings of the International Conference on Intelligent Computing and Control Systems Iciccs 2020, 2020 The impact of harmful pollutants in the air on human health is a vast area of research, preventing or controlling, and also monitoring the pollutant is the huge responsibility of any governing body. Several computing models starting from statistical and machine learning to deep learning have compared and contrasted to prove the accuracy of forecasting air quality standards until date. The level of pollutants is still not in control in several parts of the world due to various sources and reasons. This paper attempts to forecast PM2.5 pollutant which is one of the detrimental diseases triggering pollutants throughout the globe by using bidirectional long short term memory model. The proposed model accuracy is comparatively greater than the existing model by evaluating the following error estimation metrics Root mean square error = 9.86, mean absolute error = 7.53, and symmetric mean absolute percentage error = 0.1664.
Presence or absence of TB classification, case control study and analysing the positive association with air pollution Journal of Advanced Research in Dynamical and Control Systems, 2019
Assessment of dysarthria speech disorder using MFCC and Hidden Markov model Journal of Advanced Research in Dynamical and Control Systems, 2018
