Publications, Research, and Certifications
A collection of my professional credentials, ongoing research, and published work.
Publications
Revisiting the Hough Transform: Lightweight Road Detection with Agentic Intelligence
A lightweight hybrid framework combining Hough Transform with agentic semantic validation for road detection. Achieves improved adaptability and stability without dependency on deep neural networks, labeled datasets, or GPU acceleration, while maintaining lower computational overhead than deep learning alternatives.
Conference Name: CIMA 2026 - NIT Puducherry, India
View DocumentAnalysis of Random Forest, CNN, and Bidirectional LSTM for Intrusion Detection on the NSL-KDD Dataset
This paper presents a comparative study of Random Forest, 1D CNN, and BiLSTM models for intrusion detection using the NSL-KDD dataset. With a unified preprocessing pipeline and focus on challenging attack classes such as R2L and U2R, the study highlights the strengths of each model and demonstrates that Random Forest provides the most consistent and reliable performance.
Conference Name: IDSCS 2025
View DocumentExploring MobileNetV2 and SqueezeNet for Efficient Fruit Freshness Detection in Agricultural Systems
This paper explores the use of MobileNetV2 and SqueezeNet for classifying fruit freshness in agricultural systems. By leveraging lightweight CNN architectures and Grad-CAM for interpretability, the study achieves highly efficient and accurate classification suitable for real-time deployment.
Conference Name: ICATEST 2025
View DocumentA Comparative Study of Filter, Wrapper and Embedded Feature Selection Techniques for Intrusion Detection on the NSL-KDD Dataset.
This paper presents a systematic study on intrusion detection using the NSL-KDD dataset, comparing filter, wrapper, and embedded feature selection methods with XGBoost as the primary classifier. The results show that effective feature selection and class balancing techniques significantly improve model performance and overall detection capability.
Conference Name: IJCACI 2025
View DocumentA Hybrid Approach for Intrusion Detection Using Feature Selection and Class Balancing on the NSL-KDD Dataset
This paper provides a comparative evaluation of feature selection and class balancing strategies on the NSL-KDD dataset using Random Forest and XGBoost. The findings highlight that combining feature selection with balanced sampling leads to stronger detection of minority attack classes and more reliable overall intrusion detection performance.
Conference Name: CIS 2025
View DocumentHybrid Stock Movement Prediction Using Market Indicators and FinBERT-Based Sentiment Analysis
This paper proposes a hybrid stock prediction model that combines technical indicators with FinBERT-based sentiment analysis of financial news to predict short-term stock movements. Using XGBoost and explainable AI techniques like SHAP, the model achieves 66.7% prediction accuracy while maintaining interpretability. The study demonstrates how integrating market sentiment with quantitative indicators can improve financial forecasting and decision-making.
Conference Name: AIR 2026
View DocumentResearch Work
Hybrid Post Quantum Cryptography Cryptosystems: ML-KEM (Kyber) Key Establishment with AES-256 AEAD
This paper proposes a hybrid post-quantum cryptographic framework combining lattice-based key encapsulation (ML-KEM/Kyber) for key exchange with AES-256 AEAD for bulk encryption. It outlines a TLS-style handshake, transcript-bound key derivation using HKDF, and strict nonce discipline for security. The hybrid design maintains AES throughput while adding post-quantum protection against “store-now, decrypt-later” threats. Evaluations confirm minimal handshake overhead and near-baseline performance, aligning with NIST PQC transition guidance.
College Research Work - Cryptography and Network Security
View DocumentBenchmarking TF-IDF, Word2Vec-LSTM, and Legal-BERT for Automated Legal Clause Classification
This study evaluates three NLP approaches—TF-IDF with SVM, Word2Vec-LSTM, and Legal-BERT—for classifying legal contract clauses. Through comparative analysis, it shows how transformer-based models capture domain-specific semantics more effectively than statistical and sequential embedding methods. Legal-BERT achieved ~93.7% accuracy and the highest precision and F1-scores, highlighting its superiority in handling complex legal language. The results demonstrate the value of contextual embeddings in building reliable and scalable LegalTech systems.
College Research Work – Natural Language Processing
View DocumentCertifications
AWS Academy Graduate - Cloud Architecting - Training Badge
Completed AWS Academy Cloud Architecting training, gaining expertise in designing and deploying scalable, secure, and cost-effective cloud solutions on AWS. Skilled in applying AWS best practices using the Well-Architected Framework.
AWS Academy
View DocumentMastering Centralized Application Configuration in Azure: A Strategic Approach.
The workshop focused on key strategies for managing application configurations within the Microsoft Azure cloud environment. Key topics covered included best practices for secure configuration storage, with a specific focus on utilizing Azure App Configuration and Azure Key Vault.
NPTEL
View Document