Who's reading?

Choose a view to put what you are looking for first. The page and the menu re-prioritise around it, and a margin note always says why.

Who's reading?

Research

Research that adapts.

Three themes: Human factors in cyber security; Adaptive and explainable AI in education; AI for assessment and content. Projects, publications and my doctoral proposal, all on this page.

01 · Statement

Research statement

Most successful cyber attacks start with a person, not a machine. In Sri Lanka's Northern Province, students are coming online faster than schools can teach them to stay safe, and generic awareness programmes rarely change behaviour. My research asks how AI can adapt cyber-safety education to each learner — and explain its reasoning so that students and teachers can trust it. I build and evaluate systems at this intersection: CyberSafe, an AI-driven framework for predicting and reducing students' cybercrime vulnerability; AKARAM, an AI-assisted learning-difficulty screening platform; and an artificial-neural-network system that personalises secondary mathematics. I am preparing for doctoral study to take this work further.

Children are now one in three internet users worldwide, and each year more than 300 million experience technology-facilitated sexual exploitation or abuse. Phishing, online scams and cyberbullying succeed by exploiting people rather than machines. Developing countries have far less evidence and capacity to respond, and Sri Lanka shows the gap: in national research by the University of Colombo, 28% of children had experienced online violence and 67% would not confide in their teachers. The Northern Province, where I live and work, is largely missing from this evidence.

My research sits where artificial intelligence, the human factors of cyber security and computing education meet. I see four gaps. Sri Lanka's ICT curriculum has never been analysed for how well it covers human-factor risks. No data-driven model shows which human factors predict a student's vulnerability across the 4Cs of online risk: content, contact, conduct and contract. Most cyber-safety interventions for children are static and rarely evaluated rigorously. And while generative AI can create realistic threat scenarios, no framework yet combines it with learner modelling, explainability and local safeguards.

My work so far builds towards closing them. CyberSafe, presented at NPAIRC 2026, is an AI-driven framework for predicting and reducing students' cybercrime vulnerability. My MSc research used an artificial neural network to personalise mathematics for secondary students. AKARAM screens primary children in Tamil for possible learning difficulties, and LankaMCQ-Gen helps teachers build curriculum-aligned assessment. Alongside them, I study cyber complaints in Jaffna and the web security of Sri Lankan institutions, and my responsible-disclosure work has been acknowledged by more than 50 organisations.

For doctoral study I propose CyberSafe-AI: an explainable, adaptive generative AI framework that personalises curriculum-aligned cyber security learning for secondary school students, designed, built and evaluated with schools in the Northern Province. The aim is a model that tells teachers which students are most at risk and why, and a learning system that adapts to each student while explaining its choices, so that students and teachers can trust it.

02 · Themes

Three themes.

  1. Human factors in cyber security

    Why students and everyday users fall for social engineering, and what that looks like in the Northern Province of Sri Lanka.

  2. Adaptive and explainable AI in education

    Systems that personalise learning to each student and explain their decisions: personalised mathematics with neural networks, learning-difficulty screening with AKARAM, and adaptive cyber-safety feedback with CyberSafe.

  3. AI for assessment and content

    Generating and evaluating assessment material with AI, starting with multiple-choice papers for Grades 6–11 in LankaMCQ-Gen.

03 · Projects

The projects.

Every project has a simple summary and a technical one; switch with the reading level control.

Active, 2025–present

CyberSafe

CyberSafe is an AI-driven, adaptive framework for predictive cybercrime vulnerability assessment among Sri Lankan secondary school students (Grades 9–13). It models human-factor risk indicators from learner responses and behaviour, predicts individual vulnerability profiles, and sequences personalised, explainable interventions. [CONFIRM: model type, features, sample, evaluation metrics from the paper]

Many students in Sri Lanka use phones and social media every day but have never been taught how scams, fake links and online grooming work. CyberSafe uses AI to work out which students are most at risk and then gives each of them short lessons that fit what they already know — and explains why it chose them.

Themes: Human factors in security, Adaptive AI, Explainable AI, Cyber security education

Active, [CONFIRM]

AKARAM

AKARAM is a Tamil-medium learning-difficulty screening platform for primary-grade children. Teachers and parents answer structured questionnaires across eight areas, including reading, writing, mathematics, language, memory and attention. Results flag possible specific learning disorders (dyslexia, dysgraphia, dyscalculia), ADHD and dysarthria, each with suggested support activities in Tamil. It screens and does not diagnose. [CONFIRM: how results are scored (rules or model), validation approach, deployment stage]

Some children struggle with reading, writing, maths, speech or attention for reasons that are never spotted early. AKARAM (அகரம்) is a Tamil-language tool that helps teachers and parents of primary-school children notice the signs early, and suggests activities to support each child. It is a screening tool, not a diagnosis.

Open AKARAM · akaram.hexad.org.uk

Themes: Adaptive AI, Learning analytics

Published, 2022–2026

Personalised mathematics with artificial neural networks

An artificial neural network-based personalised system to enhance mathematics education for secondary students in Sri Lanka. [CONFIRM: input features, architecture, dataset size and source, accuracy/F1, how predictions drive personalisation]

Every student learns maths differently. This project trained a neural network to predict where each secondary student is likely to struggle, so lessons and practice can be personalised for them.

Themes: Adaptive AI, Learning analytics

Active, [CONFIRM]

LankaMCQ-Gen

LankaMCQ-Gen generates multiple-choice papers for Grades 6–11 of the Sri Lankan curriculum, grounded in syllabus units and textbook text, and can also assemble papers from past papers and teacher-approved items. Questions can be produced or translated into Tamil, Sinhala and English. Teachers accept, edit or swap each item; approved questions feed a teacher-verified bank that informs later generation. [CONFIRM: generation model, distractor method, evaluation (expert rating, difficulty and discrimination indices)]

Writing a good multiple-choice paper takes teachers hours. LankaMCQ-Gen builds MCQ papers for Grades 6–11 from the school textbooks and syllabus, in Tamil, Sinhala or English. Teachers check, edit and approve the questions, and approved questions go into a shared, verified question bank. It is being tested with teachers.

Open LankaMCQ-Gen · mcq.hexad.org.uk

Themes: AI for assessment, Adaptive AI

In preparation, 2026–2027

Cyber complaints in Jaffna District

A descriptive study of cyber complaint statistics across Jaffna District police stations — complaints lodged, solved and pending, disaggregated by age category — with a planned extended study with the National Police Academy in 2027. Data will be published only in aggregate and with approval.

How many cyber-related complaints reach police stations in Jaffna, who makes them, and how many are solved? This study collects that data so that awareness programmes can target the right age groups.

Themes: Cybercrime data, Human factors in security

In preparation, 2026

Web security of Sri Lankan state universities

A census-based, cross-sectional passive measurement study of the main websites of all 17 state universities under the University Grants Commission (.ac.lk). It assesses TLS configuration, HTTP security headers, email authentication (SPF, DMARC, DNSSEC), CMS and server software currency, potential exposure to known CVEs (NVD CPE matching), and mobile and desktop performance (Lighthouse and Chrome UX Report), with a review of publicly reported incidents and Spearman correlation between security and performance. Non-intrusive methods only, in line with the Computer Crime Act No. 24 of 2007. Submitted to ICARC 2027; results will be shared here after publication and after the universities and Sri Lanka CERT have been informed.

University websites carry admissions notices, exam results and research, yet no one had checked how secure or how fast the websites of Sri Lanka's state universities are. This study measures all 17, using only the checks any visitor's browser could make, and recommends a simple security baseline for the whole sector.

Themes: Web security, Human factors in security

Planned, 2026

Securing low-cost IoT energy meters

Extends an ESP32/ESP8266-based energy meter with data collection and monitoring (voltage and current sensing, Blynk app, cloud logging) with a cyber-physical security contribution — device authentication, secure telemetry and tamper detection — targeting an IEEE cyber-physical and IoT security venue.

Cheap smart energy meters built on hobby boards often send data with no protection at all. This project adds practical security to one such meter and tests how well it holds up.

Themes: IoT security

Try the live tools.

  • akaram.hexad.org.uk

    AKARAM

    Some children struggle with reading, writing, maths, speech or attention for reasons that are never spotted early. AKARAM (அகரம்) is a Tamil-language tool that helps teachers and parents of primary-school children notice the signs early, and suggests activities to support each child. It is a screening tool, not a diagnosis.

  • mcq.hexad.org.uk

    LankaMCQ-Gen

    Writing a good multiple-choice paper takes teachers hours. LankaMCQ-Gen builds MCQ papers for Grades 6–11 from the school textbooks and syllabus, in Tamil, Sinhala or English. Teachers check, edit and approve the questions, and approved questions go into a shared, verified question bank. It is being tested with teachers.

Research mapSee how the projects, papers and posts connect.

04 · Publications

Papers & talks.

Download all as BibTeX

Each entry has a citation you can copy. For the full record, Download CV (PDF, 186 KB).

Conference papers

Babiharan, S., & Dilukshika, S. (2026). CyberSafe: An AI-Driven Ludic-Pedagogical Framework for Predictive Cybercrime Vulnerability Assessment Among Sri Lankan School Students. National Police Academy International Research Conference (NPAIRC 2026).

Theses

05 · Doctoral proposal

The proposal.

Working titleAn Explainable, Adaptive Generative Artificial Intelligence Framework for Personalised Cyber Security Education: Addressing Human-Factor Vulnerabilities Among Secondary School Students

Abstract

Children have become prominent targets of cybercrime: each year more than 300 million children worldwide experience technology-facilitated sexual exploitation or abuse. Attacks such as phishing, online scams and cyberbullying exploit human rather than technical weaknesses, in developed and developing countries alike, yet developing countries have far less evidence and capacity to respond. Sri Lanka, only recently reclassified as an upper-middle-income country, shows this gap, as does its Northern Province, which generates just 4.5% of national GDP. At the intersection of artificial intelligence, the human factors of cyber security and computing education, this research will first build a data-driven evidence base on the threats facing secondary students in the Northern Province, focusing on Jaffna District: their underlying human-factor vulnerabilities, and how far the national ICT curriculum covers them. It will then design, develop and evaluate CyberSafe-AI, a framework that integrates explainable machine learning, adaptive learner modelling and retrieval-augmented generative AI to deliver personalised, curriculum-aligned cyber security learning. Following design science research, the framework will be evaluated technically and pedagogically in Northern Province schools.

Research questions

RQ1: What cyber threats do secondary students in the Northern Province face, which human factors predict their vulnerability, and how far does the national ICT curriculum address them?

RQ2: How can explainable, adaptive generative AI deliver personalised, curriculum-aligned cyber security learning safely?

RQ3: How effectively does the framework improve students' cyber security knowledge, self-efficacy and protective behaviour?

Approach

Design science research in three phases.

Phase 1 (RQ1): a PRISMA 2020 systematic review; analysis of the GCE O/L and A/L ICT syllabus; a survey of about 2,500 students in Grades 9–13 across all education zones of the Northern Province. Logistic regression, random forest, XGBoost and a multilayer perceptron are compared with stratified 10-fold cross-validation, and the best model is explained with SHAP, with fairness checked across gender, school type and location.

Phase 2 (RQ2): CyberSafe-AI combines the Phase 1 learner-risk model, an adaptive engine using Deep Knowledge Tracing (benchmarked against Bayesian Knowledge Tracing), and retrieval-augmented generation that grounds an open-source language model in verified local threat cases. Teachers validate every scenario; delivery is game-based, with a teacher dashboard.

Phase 3 (RQ3): a quasi-experimental pre-test, post-test and delayed post-test study in about 12 schools (around 600 students), analysed with ANCOVA, learning analytics and student focus groups.

Contribution

The first evidence-based model of human-factor cyber vulnerability for secondary school students in the Northern Province, with an assessment of how the ICT curriculum deals with it.

A new open-source, explainable and adaptive generative AI framework, with evidence of its effectiveness. The framework is designed to be context-independent, so it can be applied in other regions, such as England.

Feasibility and ethics

Planned over four years of full-time registration: review, ethics and instruments in year 1; curriculum analysis, survey and modelling in years 1–2; building CyberSafe-AI in years 2–3; school evaluation in years 3–4; writing up in year 4.

Ethical approval from the university's research ethics committee and Sri Lanka's Ministry of Education, with parental consent, child assent and safeguarding procedures. Data are anonymised, encrypted and handled under the UK GDPR and Sri Lanka's Personal Data Protection Act No. 9 of 2022. Self-report bias, non-random assignment and generative AI errors are mitigated through triangulation, baseline covariates and teacher review.

Supervisors: the full proposal, with references, is available on request. Ask for it here.

Back to top

Keyboard shortcuts

⌘ K / Ctrl K
Search the site and run commands
`
Open the terminal
?
Show these shortcuts
Esc
Close any panel or menu
Tab / Shift Tab
Move between links and buttons

Terminal

Type help to see what you can do here. Press Escape or type exit to close.