Research archive

Papers & publications.

Research spanning geometric and signed-network security, causal cyber defense, semi-supervised learning, geospatial carbon estimation, natural language processing, and intelligent infrastructure.

ORCID ↗Google Scholar ↗ResearchGate ↗

9archived works

10Scholar citations

1Scholar h-index

Scholar snapshot: 27 October 2026. Seven indexed articles; i10-index: 0. This archive also includes two earlier reports/preprints. View live Scholar metrics ↗

Research themes

Causal cyber defenseMalicious URL detectionSemi-supervised learningSigned-network securityNLP & embeddingsIntelligent infrastructureGeometric threat detectionGeospatial carbon estimation

How the labels work: “Preprint” means a manuscript is publicly available before a peer-reviewed venue is recorded. “Peer reviewed” describes the scholarly evaluation process, while “conference paper” describes where the work was published.

2026

arXiv preprint · Submitted 20 October 2026 · Revised 23 October 2026 · arXiv:2610.27675v2

CyberHull: Deterministic Malicious URL Isolation via Convex Hull

Avijit Gayen · Aniket Ghosal · Durba Chakraborty · Sayan Mondal · Md Abdul Aziz Al Aman · Angshuman Jana

CyberHull builds convex boundaries around legitimate web behaviour using content entropy, semantic features, and hyperlink topology. Deterministic containment tests identify URLs that fall outside the learned regions.

ProblemDetect evolving malicious URLs with inspectable, computationally lightweight decisions.

ApproachQuickHull-based geometric vaults across complementary feature spaces.

Reported result92% accuracy and 0.92 F1 in the v2 preprint’s benchmark; broader traffic validation remains future work.

Cite this work Download BibTeX ↓

View research page → arXiv ↗PDF (v2) ↗DOI ↗Code & data ↗

arXiv preprint

2026

arXiv preprint · cs.CR · Submitted 19 August 2026 · arXiv:2608.19190

SiNMULI: Novel Signed Network Approach for Malicious URL Identification

Avijit Gayen · Sayan Mondal · Angshuman Jana

SiNMULI models websites as nodes and hyperlinks as directed trust-or-distrust edges. It applies social-balance reasoning to infer missing edge signs, then identifies unseen domains through a 51% majority rule over their incoming links.

ProblemBlacklists and static URL features can struggle with new, fast-changing, or deliberately obfuscated threats.

ApproachA directed signed hyperlink graph with balance-based edge inference and rule-based node classification.

Reported result99.89% accuracy and 99.80% F1 on a graph of 48,205 nodes and 384,592 directed edges.

Cite this work Download BibTeX ↓

arXiv ↗PDF ↗DOI ↗Code & data ↗View research page →Google Scholar ↗

arXiv preprint

2026

IEEE Region 10 Symposium (TENSYMP) 2026 · Penang, Malaysia · pp. 1–5 · DOI 10.1109/TENSYMP68909.2026.11704039

ASTriA: Adaptive Semi-Supervised Tri-Learning Approach for Scalable Malicious URL Detection

Avijit Gayen · Sayan Mondal · Bhanu Bariar · Khokan Mondal · Angshuman Jana

ASTriA is a lightweight semi-supervised approach to malicious URL detection that iteratively expands limited labeled data with high-confidence pseudo-labels from a larger unlabeled pool.

ProblemBuilding large, current, labeled URL datasets is costly, while blacklist-only defenses struggle with newly generated URLs.

ApproachTri-learning with iterative pseudo-labeling and lightweight lexical and host-based features.

FocusScalable and interpretable detection using classifiers such as Random Forest and k-Nearest Neighbors.

Cite this work Download BibTeX ↓

View research page → DOI ↗IEEE Xplore ↗Google Scholar ↗

Peer-reviewed conference paper

2022

IJARCCE · Volume 11, issue 3 · March 2022 · pp. 196–203 · DOI 10.17148/IJARCCE.2022.11336

Carbon loss estimation: a case study of Little Andaman development plan

Sujit Raha · Avijit Chakraborty · Purbita Chatterjee · Tanmoy Chakraborty · Sayan Mondal

A geospatial estimation study connecting Bhuvan land-cover data with forest carbon inventories to assess carbon stock potentially affected by the Little Andaman development scenario described in 2022.

DataResourcesat-2 LISS-III land-cover data from 2015–16 and the India State of Forest Report 2019.

MethodQGIS processing and forest-class area estimates across five carbon pools.

InterpretationA historical stock-loss estimate; the detailed page explains the source’s unit and study-area inconsistencies.

Cite this work Download BibTeX ↓

View research page → DOI ↗Publisher ↗Full text ↗

Journal article

2026

ACM AsiaCCS ’26 · Bangalore, India · 1–5 June 2026 · pp. 1919–1921 · DOI 10.1145/3779208.3804893

POSTER—CAIRN: Causal Active Inference for Real-Time Cyber Attack Diagnosis and Response

Sayan Mondal · Avijit Gayen · Angshuman Jana

CAIRN models enterprise telemetry as a probabilistic causal evidence graph and performs real-time Bayesian inference over competing attack explanations. When passive evidence is insufficient, it selects low-risk interventions—such as focused probes, enhanced logging, decoys, or scoped isolation—to generate discriminative observations.

ProblemCorrelation-heavy IDS and SIEM workflows generate alerts without enough causal clarity.

ApproachEvidence graphs, Bayesian diagnosis, GenAI-assisted CTI extraction, and an online intervention loop.

ContributionAn infer-and-verify defense workflow that reduces false alarms and confirms attacks faster than a passive IDS baseline in testbed evaluation.

Cite this work Download BibTeX ↓

DOI ↗ACM Digital Library ↗DBLP ↗View research page →Google Scholar ↗

Peer-reviewed poster paper

2026

ISEC 2026 · Research Papers Track · Short paper · DOI 10.1145/3796563.3796574

EntroURL-Bench: A Feature-Rich Benchmark Dataset for Trustworthy Malicious URL Detection

Avijit Gayen · Sayan Mondal · Khokan Mondal · Angshuman Jana

Malicious-URL classifiers are only as reliable as their evaluation data. EntroURL-Bench addresses common limitations in public datasets—including age, shallow lexical representations, and weak coverage of modern attacks—with a feature-rich benchmark intended for reproducible, trustworthy evaluation.

ProblemExisting URL datasets can be outdated, narrow, or poorly suited to modern feature-rich detection.

FocusPhishing, malware, and malicious-web threat analysis using richer URL representations.

ValueA consistent benchmark for comparing detection methods and studying model trustworthiness.

Cite this work Download BibTeX ↓

DOI ↗ACM Digital Library ↗Conference record ↗View research page →Google Scholar ↗ResearchGate ↗

Peer-reviewed conference paper

2020

ICCECE 2020 · Conference paper · DOI 10.1109/ICCECE48148.2020.9223092

Global Positioning System Based Automated Railway Level Crossing

Shankha Banerjee · Sayan Mondal · Amartya Chakraborty · Suvendu Chattaraj

The paper proposes a GPS-based approach for detecting an approaching train and controlling level-crossing gates automatically. By avoiding additional trackside detection hardware, the design aims to reduce vandalism risk, gate-operation delay, and human coordination errors while improving passenger safety.

ProblemManual crossings depend on timely communication and gatekeeper action, creating avoidable risk.

ApproachTrain location from GPS drives automated opening and closing decisions.

GoalSafer, faster crossing operation with less dedicated wayside hardware.

Cite this work Download BibTeX ↓

View research page → DOI ↗Full text ↗Google Scholar ↗

Peer-reviewed conference paper

2020

Technical report · March 2020 · DOI not assigned

Some Observations and Problems Explored on Dimensionality Reduction of Text Documents

Somyajit Chakraborty · Sayan Mondal

An exploratory study of problems encountered while reducing the dimensionality of text representations during document embedding. The report records practical observations and discusses possible solutions before the authors’ broader research proposal.

DomainNatural language processing and document representation.

QuestionHow can dimensionality be reduced while retaining useful structure in embedded text?

ContributionA focused analysis of observed failure modes and candidate remedies.

Cite this work Download BibTeX ↓

View research page → ResearchGate ↗Google Scholar ↗

Technical report

2017

Preprint · DOI 10.13140/RG.2.2.13803.16167

Smart Street Light Control System Using an Algorithm and Image Processing

Sayan Mondal

A camera-based control concept that adjusts street-light configuration according to observed road workload or vehicle density. The work connects image processing with adaptive illumination to reduce unnecessary energy use while maintaining road visibility and security.

ProblemStatic public lighting consumes energy regardless of actual road demand.

ApproachCamera observations and image processing estimate activity and drive lighting control.

GoalLower municipal energy use without compromising safety.

Cite this work Download BibTeX ↓

View research page → DOI ↗Full text ↗Google Scholar ↗

Preprint

Cite & follow

Track the research.

Every publication above includes a BibTeX download and a one-click citation-copy action. Google Scholar remains available for live citation counts and additional export formats.

Open Scholar profile ↗