Prospective doctoral researcherUK · EU · Remote collaboration

Explainable AI · Computer vision

Research that earns trust.

I am Ashraf Hussan Babor, a prospective PhD applicant with an MSc in IT with Data Analytics (Distinction), seeking a supervisor and fully funded opportunity in explainable AI, trustworthy machine learning or computer vision.

Discuss supervision View IEEE paper

2completed theses

1peer-reviewed IEEE paper

UK / EUfunded PhD search

01 / Research direction

From model performance to dependable decisions.

My BSc research built and compared deep-learning pipelines for a real-world visual recognition problem. My MSc work then examined how predictive models explain their decisions in a high-stakes cardiovascular-risk setting.

These projects converge on the doctoral question I want to pursue: how can we know when a model and its explanation remain reliable beyond the data on which they were developed? I am interested in explanation fidelity, robustness under distribution shift, comparative evaluation and practical auditing protocols.

Alongside research, founding Dashient has given me experience translating uncertain data into inspectable systems. I bring that applied perspective to work intended for real practitioners, not only benchmark performance.

02 / Qualifications

Postgraduate qualification

MSc IT with Data Analytics — Distinction

University of the West of Scotland

Thesis: “Building Trustworthy AI for Cardiovascular Disease Prediction: Quantifying Explainability and Robustness Across Machine Learning and Deep Learning Models.” Supervised by Dr Martins Olaleye.

Peer-reviewed publication · BSc thesis output

Deep Learning Techniques for Bangladeshi Coin Detection and Automated Counting System

IEEE ICAEEE 2024 · pp. 1–6 · DOI 10.1109/ICAEEE62219.2024.10561714

A comparative study of VGG16, VGG19 and EfficientNetB0 spanning dataset preparation, augmentation, model training, evaluation and an automated counting system.

Professional certification

Microsoft Certified: Azure Fundamentals

AZ-900

Foundational knowledge of cloud concepts, services, governance and the operational context around deployed data systems.

03 / Research projects

Research A · Published undergraduate research · IEEE ICAEEE 2024

Bangladeshi Coin Currency Detection and Recognition

Objective: Recognise Bangladeshi coins from images and support automated counting.
Method: Comparative study of transfer-learning CNNs with dataset preparation, augmentation, training and evaluation.
Models: VGG16 · VGG19 · EfficientNet-B0
Finding: Reported average accuracy of approximately 97% across the compared models.

Read on IEEE Xplore (DOI 10.1109/ICAEEE62219.2024.10561714)

Research B · MSc dissertation · University of the West of Scotland

Building Trustworthy AI for Cardiovascular Disease Prediction: Quantifying Explainability and Robustness Across Machine Learning and Deep Learning Models

Data: UCI Cleveland cohort (303 patients) for development; Hungarian and VA Long Beach cohorts for independent external validation.
Models: Logistic Regression · Explainable Boosting Machine · Random Forest · XGBoost · Feedforward Neural Network
Methods: Repeated stratified 10-fold cross-validation × 5, SMOTE, SHAP and LIME explanation evaluation, measurement-noise robustness testing and a measurement-based Trustworthiness Score.

Internal performance — Cleveland cohort, repeated stratified 10-fold CV × 5
Model	AUC	Accuracy	Recall	F1	Brier
Logistic Regression	0.912	83.7%	78.7%	81.4%	0.118
Explainable Boosting Machine	0.909	82.7%	78.7%	80.6%	0.121
Random Forest	0.911	82.6%	77.8%	80.3%	0.124
XGBoost	0.897	81.3%	76.4%	78.8%	0.134
Feedforward Neural Network	0.892	81.2%	76.8%	78.6%	0.132

External validation — AUC under dataset shift
Model	Hungarian	VA Long Beach
Logistic Regression	0.878	0.739
Explainable Boosting Machine	0.880	0.732
Random Forest	0.889	0.744
XGBoost	0.869	0.741
Feedforward Neural Network	0.885	0.731

Trustworthiness Score (primary 10/30/30/30 weighting) and explanation metrics
Model	Trust score	SHAP stability	SHAP–LIME agreement
Random Forest	89.8	0.81	0.81
Explainable Boosting Machine	86.4	0.61	0.79
XGBoost	86.4	0.75	0.68
Feedforward Neural Network	86.4	0.76	0.76
Logistic Regression	85.0	0.65	0.72

No model differed significantly in AUC from Logistic Regression internally (all p ≥ 0.13).
All models lost discrimination under external shift: AUC ≈ 0.87–0.89 (Hungarian), 0.73–0.74 (VA Long Beach).
Measurement-noise prediction-flip rates ranged from 0.4% to 2.7%.
Random Forest ranked first in 96.9% of 20,000 randomly sampled score weightings.
Conclusion: predictive performance alone is insufficient to judge trustworthy clinical AI.

Academic research only — not clinically deployed or clinically validated.

04 / Research interests

Explainable & trustworthy AI

Methods that make model decisions intelligible, testable and useful to practitioners.

Computer vision & deep learning

Robust recognition systems, comparative architectures and evaluation beyond headline accuracy.

Explanation reliability

Testing SHAP, LIME and related methods for stability, fidelity and behaviour under distribution shift.

Model auditing

Protocols that reveal when a model—or its explanation—has stopped being dependable.

Causal attribution

Moving from descriptive channel credit toward incrementality and defensible causal measurement.

Applied decision support

Responsible systems that connect rigorous modelling to real clinical and commercial decisions.

Current search

Seeking the right supervisor and funded route.

Explainable or trustworthy machine learning
Deep learning and computer vision
Robustness, model auditing or distribution shift
UK or EU studentship, scholarship or CDT
Open to an initial paper collaboration

Research enquiry

Start a conversation.

If my background could fit your lab, programme or funded project, use the dedicated enquiry page to share your background and a possible research topic.

Or email ashrafhussanmailofficial@gmail.com directly

A structured introduction

Share the supervision context.

The enquiry asks about your institution, research background, proposed topic, funding context and preferred start period.

Open supervision enquiry