Research Focus — Trustworthy & Explainable Machine Learning
I study how machine learning models fail quietly — across shifting Android malware domains and opaque decision boundaries. My projects and research work include Android malware attribution, model interpretability, cross-domain generalization, and healthcare AI applications.
01 — About
I completed my BSc in Software Engineering from the Department of Computer Science and Engineering at the Islamic University of Technology (IUT). During undergraduate studies, my research focused on the efficiency of using Android permission requests as features, combined with supervised learning, to classify mobile applications as benign or malicious. After graduation, I spent some time abroad and traveled across North America, experiences that broadened my global perspective and professional outlook.
I have recently completed an MSc in Computer Science and Engineering with a concentration in intelligent systems engineering from the Department of Electrical and Computer Engineering at North South University. My research interests span explainable AI, cybersecurity, and trustworthy AI. My work focuses on developing robust and interpretable machine learning methods, with applications in mobile security and healthcare. Beyond academics, I am deeply interested in history, international relations, theology, and jurisprudence, which continue to shape my worldview and analytical thinking.
M.Sc. in Computer Science & Engineering
Concentration: Intelligent Systems Engineering
Graduated September 2026
North South University · CGPA 3.69/4.00 ·
B.Sc. in Software Engineering
Graduated May 2022
Islamic University of Technology · CGPA 3.12/4.00 (Upper-division 3.51/4.00)
02 — Research
Every entry below is stamped by status — the way it actually moves through the pipeline, from submission to print.
Explores CNN fusion architectures for diabetic retinopathy screening, weighing diagnostic accuracy against computational efficiency for real-world clinical deployment.
IEEE 28th Int'l Conference on Computer and Information Technology (ICCIT), Cox's Bazar
Examines permission-based feature sets for Android malware detection, benchmarking classical machine learning approaches for practical deployability.
IEEE 15th Int'l Conference on Security of Information and Networks (SIN), Sousse
Investigates how pseudo-labeling gains vary by classifier family in semi-supervised malware attribution, isolating which model types benefit most from unlabeled data.
IEEE Int'l Conference on Signal Processing, Information, Communication and Systems (SPICSCON)
Proposes an intrinsic-motivation exploration strategy for personalized federated reinforcement learning, improving client-level adaptation without centralizing data.
IEEE Int'l Conference on Signal Processing, Information, Communication and Systems (SPICSCON)
Diagnoses the sources of performance decay when permission-based malware detectors are transferred across datasets, and proposes mitigation strategies for robustness.
Under review — Pervasive and Mobile Computing
Applies quantum learning techniques to automated sleep-stage classification from polysomnographic recordings, comparing against classical baselines.
Under review — IEEE Transactions on Emerging Topics in Computational Intelligence
A 25-year longitudinal analysis of Bangladeshi air quality data, characterizing long-term trends, seasonal cycles, and spatial pollution regimes.
IEEE 11th Asia-Pacific Conference on Computer Science and Data Engineering (CSDE)
This paper introduces CAFE-AQI, a coverage-aware forecasting framework that accurately predicts next-day maximum air quality and provides unhealthy-air warnings in Bangladesh by accounting for data gaps and duplicate city trajectories.
Under review — IEEE 29th International Conference on Computer and Information Technology (ICCIT).
03 — Presentations
Classifier-Dependent Benefits of Pseudo-Labeling for Semi-Supervised Android Malware Attribution
IEEE 5th SPICSCON · Natore, Bangladesh
Balancing Accuracy and Efficiency: CNN Fusion Models for Diabetic Retinopathy Screening
IEEE 28th ICCIT · Cox's Bazar, Bangladesh
A Study of Permission-Based Malware Detection Using Machine Learning
IEEE 15th SIN · Sousse, Tunisia
04 — Toolkit
The stack I reach for when moving from a research question to a working, explainable model.
Ensemble Learning, Semi-Supervised Learning, Deep Learning, CNNs, Transfer Learning, Hyperparameter Tuning, Cross-Validation
SHAP, LIME, Feature Attribution Analysis
Cross-Domain Evaluation, Ablation Studies, Model Generalization Analysis, Dataset Harmonization
Feature Selection, Cross-Dataset Feature Alignment, Correlation Analysis, Imbalanced Data Handling
Android Malware Detection, Permission-Based Analysis, Static & Dynamic Feature Analysis
Image Classification, Medical Image Analysis, Data Augmentation, CNN Fusion
PyTorch, scikit-learn, NumPy, Pandas, Matplotlib, Seaborn
Software Quality Assurance, Mobile & Web Development, UI/UX Design (Figma, Adobe XD)
05 — Contact
Reach out about collaborations in explainable AI, malware detection, healthcare ML, or graduate opportunities — I read every message.
B-120, Banasree, Dhaka, Bangladesh