Communities in Avinashilingam Institute for Home Science and Higher Education for Women - AULIB-IR Central Library
Select a community to browse its collections.
Recent Submissions
Classification of Diabetic Retinopathy Stages Using Deep Learning Architectures
(Avinashilingam, 2025-08) Santhiya Lakshmi K; Guide - Dr. B. SARGUNAM
Diabetic Retinopathy (DR) is a progressive microvascular complication of diabetes and a leading cause of vision impairment and blindness worldwide. Early detection and accurate classification of DR are essential for timely medical intervention and prevention of severe visual deterioration. Traditional diagnostic approaches rely on manual grading by ophthalmologists, which is time-consuming, labor-intensive, and susceptible to inter- observer variability. Recent advancements in Deep Learning (DL) have demonstrated significant potential in automating DR classification; however, existing models face challenges related to overfitting, computational complexity, and generalization across diverse datasets. This research aims to develop a robust and computationally efficient DR classification system capable of accurately detecting all four stages of DR like normal, mild, moderate and severe while addressing these challenges. The proposed methodology follows a three step approach to enhance model performance and adaptability. In the first phase, Transfer Learning (TL) is employed to fine-tune state-of-the-art pretrained models, particularly EfficientNetV2L, using the Asia Pacific Tele-Ophthalmology Society (APTOS) dataset. The objective is to utilize deep feature extraction capabilities while ensuring improved generalization. Data augmentation and scaling techniques are incorporated to mitigate overfitting and enhance model robustness. In the second phase, a custom Convolutional Neural Network (CNN) architecture is designed to optimize feature extraction and computational efficiency. The architecture incorporates structured convolutional layers, batch normalization, and dropout mechanisms to enhance learning stability while reducing the risk of overfitting. In the third phase, a hybrid model is developed by integrating the EfficientNetV2L with the custom CNN architecture, thereby utilising the strengths of both architectures. Extensive hyperparameter tuning is performed, including learning rate scheduling, dropout regularization, batch size optimization, and adaptive optimization using Adam optimizer, to ensure efficient training and convergence. The hybrid model is designed to provide a balance between high classification accuracy and computational feasibility, facilitating its deployment in clinical applications.The proposed system is evaluated using key performance metrics, including accuracy, precision, recall, F1-score, and computational complexity (trainable parameters).The hybrid model (EfficientNetV2L + Custom CNN Model) achieves the highest classification accuracy of 94%, with a precision of 94%, recall of 93%, and F1-score of 94%, utilizing 441,087 trainable parameters. Comparative analysis with benchmark models demonstrates the superiority of the proposed approach, with EfficientNetV2L achieving 92% accuracy, the Custom CNN Model reaching 88% accuracy, while conventional architectures such as VGG16 (85%), ResNet-50 (87%), InceptionV3 (89%), and DenseNet121 (90%) exhibit relatively lower performance. These findings confirm the effectiveness of the hybrid approach in achieving state-of-the-art classification accuracy while maintaining computational performance. The developed hybrid model is evaluated against an existing architecture that integrates DenseNet121, Xception, and EfficientNetB3, comprising 2,361,860 trainable parameters. The existing model achieves an accuracy of 75%, precision of 73%, recall of 75%, and an F1 score of 71%. In contrast, the proposed model, with only 441,087 parameters, demonstrates superior performance achieving a 19% increase in accuracy, 21% in precision, 18% in recall, and 23% in F1 score highlighting its effectiveness despite a significantly reduced parameter count. This research contributes to the advancement of automated DR classification by integrating fine-tuned pretrained models with a lightweight custom CNN, resulting in a scalable, efficient, and clinically viable diagnostic framework. The system is validated using clinical data to ensure its applicability in practical healthcare settings, particularly in resource-constrained environments where access to high-end computational resources is limited. The proposed model has the potential to significantly enhance early DR detection, improve diagnostic consistency, and ultimately contribute to better patient outcomes in ophthalmic disease management.
Creating Awareness on Organic Waste Management Practices among Selected Rural Households
(Avinashilingam, 2024-07) Vinothini R; Guide - Manimozhi K
This study investigates the impact of a training program on organic farming practices among rural households in selected areas, aiming to enhance health, reduce environmental pollution, and promote sustainable agricultural practices. The research methodology involved conducting household surveys, implementing training sessions, and evaluating the program's outcomes. The household survey gathered demographic data and insights into current agricultural practices and organic waste management. Findings highlighted diverse farming practices and challenges such as pest attacks and diseases, exacerbated by the heavy use of chemical fertilizers and pesticides. Organic waste management practices also revealed significant gaps in disposal methods and environmental awareness among rural communities. The training program focused on educating farmers about organic farming techniques, including composting and natural pest control methods. Evaluation of the training's impact showed notable improvements in farmers' knowledge and attitudes towards organic farming. Statistical analysis indicated significant changes in knowledge scores post-training, suggesting a substantial increase in understanding organic practices among participants. The adoption of organic practices post-training was another key outcome assessed. Results demonstrated a marked increase in the adoption of composting, organic growth boosters, and natural pest and disease management methods. This adoption was supported by economic benefits derived from reduced input costs and improved crop yields, contributing to rural household prosperity. Challenges identified during the study included logistical constraints in conducting widespread training and limitations in transportation for field visits. Despite these challenges, the training program succeeded in reaching a significant number of farmers and effecting meaningful changes in agricultural practices. The findings underscore the importance of targeted training programs in promoting sustainable agriculture and improving rural livelihoods. The study contributes valuable insights into the efficacy of educational interventions in transitioning farmers towards organic farming practices. Recommendations include scaling up similar training initiatives, addressing logistical barriers, and enhancing community awareness on environmental stewardship and waste management.
Seen on the shelf
(The Hindu, 2026-01-30) Nidhi Adlakha
A paper a day
(The Hindu, 2026-02-09) Elango K
மகாகவியும் ஏ.ஐ. தொழில்நுட்பமும்
(தினமணி, 2026-02-07) தினமணி
All question papers of UG Human Development CE November 2025
(Avinashilingam, 2025-09) Avinashilingam
It is a previous year question papers of UG Human Development held during November 2025
All papers of UG BA English CE November 2025
(Avinashilingam, 2025-09) Avinashilingam
It is a previous year question papers of UG BA English held during November 2025
All question papers of UG BCBT CE November 2025
(Avinashilingam, 2025-09) Avinashilingam
It is a previous year question papers of UG BCBT held during November 2025
All question papers of UG FSN CE November 2025
(Avinashilingam, 2025-09) Avinashilingam
It is a previous year question papers of UG FSN held during November 2025
A Hybrid Machine Learning approach for Detecting Intentional and Unintentional Insider Threats with Mitigation through Behavioral Biometrics and User Profiling Mechanism
(Avinashilingam, 2025-07) Asha S; Guide - Dr. D. Shanmugapriya
Insider threat is a potential threat to an organization that results in financial and reputation losses while exposing sensitive information. Past research extensively focused on external threats, and overlooked on both intentional and unintentional insider threats.Several researchers majorly focused on detecting such insider activities but fail to mitigate both intentional and unintentional insider threats. Few challenges such as mishandling imbalanced dataset and fail to incorporate feature engineering techniques, limited mitigation strategies are encountered. This research employs a hybrid machine learning approach to identify insider threats and incorporated behavioural biometrics with user profiling to mitigate both intentional and unintentional insiders effectively. A methodology comprising of three phases is proposed. It consist of Preprocessing and Insider Detection (P&ID) in Phase I, Unintentional Insider Mitigation (UIM) in Phase
II, and Intentional Insider Mitigation (IIM) in Phase III. P&ID consist of two layers - Preprocessing, and Insider Detection. In Layer 1, log data is preprocessed using data integration, encoding and tuned the nearmiss-2 sampling technique to obtain a balanced data to diminish the class imbalance problem. In Layer 2, a hybrid B-SVM combining Support Vector Machines (SVM) and Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) is applied. It classifies users into genuine, intentional insiders, and unintentional insiders. The proposed method achieved a 99.15% detection accuracy, with a low misclassification rate of 0.85% for detecting both intentional and unintentional insider threats. Once unintentional insiders are detected, the unintentional insiders are mitigated in
UIM phase. UIM phase consist of two layers – Feature engineering, and Core behavior identification. In Layer 1, Clonal Kernel Principal Component Analysis (CKPCA) is proposed for feature engineering. CKPCA integrates population subset selection, kernel mean embedding, and dimensionality reduction to improve feature representation. These features are further analyzed using Deep Belief Networks (DBN) in Layer 2 that achieved 99.84% authentication accuracy and a 0.15% Equal Error Rate (EER) of 0.15%. This phase significantly minimizes false alarms and ensures a reliable mitigation process for unintentional insiders.
In IIM phase, the detected intentional insiders are mitigated using user profiling mechanism based on their authentication outcome. IIM phase consist of three layers – Data pre-processing, Model training and evaluation, and User profiling. In Layer 1, data pre-processing is done using label encoding and train-test split. In layer 2, Decision tree is modeled to categorize users low-risk and high-risk. In Layer 3, Low-risk users with legitimate activities are profiled into the Allowlist, while users displaying malicious intent with high-risk are placed on the Denylist. This adaptive profiling ensures that intentional threats are neutralized without affecting genuine users. The methodology was validated using two datasets namely the CERT Insider Threat Dataset and the CIC Darknet Dataset. P&ID detected 8 intentional and one unintentional insider among 250,078 daily logs using CERT Dataset. P&ID is validated with darknet dataset, detected 4,783 intentional-Darknet users and 68 unintentional- Darknet users where (VPN: 42) (Tor: 21) (NonVPN: 5) among 134,305 daily activities. UIM mitigated one unintentional insider as an intentional insider using CERT log activities. UIM mitigated 68 unintentional-Darknet users as 64 Intentional-Darknet and 4
benign users using darknet dataset. IIM profiled 57 genuine users in Allowlist and 8 intentional insiders in Denylist using CERT dataset. Using CIC Darknet dataset, the IIM profiled 5063 benign users in Allowlist and 4847 Intentional-Darknet users in Denylist. This study offers a practical and highly effective solution for insider threats in environments where user log data is analyzed. By combining hybrid machine learning models with behavioral biometrics and user profiling, the approach ensures accurate detection and mitigation of both intentional and unintentional threats. This approach can be applied in any environment where user log is prevalent.