Research& Publications
Applied machine learning research in network security and scientific computing, built on reproducible pipelines and verified baselines.
Research Focus
My research asks practical questions. How should intrusion detection systems handle severely imbalanced attack classes? Can plain neural networks act as surrogates for differential equation solvers? Every study runs on a reproducible pipeline with fixed seeds, leakage-proof splits, and metrics chosen to expose weaknesses rather than hide them.
Current Focus
ML for Network Security, Neural ODE Surrogates
Long-term Goal
Graduate research in applied ML & systems
Core Interests
Research Experience
Undergraduate Thesis Researcher (Team Lead)
United International University · Team Paradox • Advisor: Dr. Muhammad Nomani Kabir
Leading FYDP research on solving the Lorenz-1960 ODE system with optimal ANN architectures. I built a verified RK4/DOP853 ground-truth pipeline (agreement RMSE around 1.3e-11) and designed a controlled 69-run architecture and optimizer search. FYDP-I was defended in June 2026.
Independent ML Researcher
UNSW-NB15 Intrusion Detection Study • Advisor: Self-directed
Designed and ran a reproducible 18-experiment grid (binary and multiclass, three models, three imbalance strategies) on UNSW-NB15, focusing on rare attack classes like Worms (0.07%) with macro-F1, ROC-AUC, and G-Mean. The paper was submitted for peer review.
Undergraduate Teaching Assistant
United International University • Advisor: CSE Department Faculty
Helping 100+ students in Data Structures & Algorithms and Database Management courses. I run tutorial sessions and grade assignments.
Publications
Handling Class Imbalance in UNSW-NB15: Reproducible Baselines for Binary and Multiclass Intrusion Detection
Under ReviewIkramul Hasan Moral*
Submitted for peer review, 2026
An 18-experiment evaluation of class-imbalance strategies (no balancing, class weighting, SMOTE) across logistic regression, random forest, and XGBoost. It includes explicit rare-class analysis (Worms: 0.07%, Shellcode: 0.65%) using macro-F1, ROC-AUC, and G-Mean on a leakage-proof, fully reproducible pipeline.
Solving the Lorenz ODE System Using Optimal ANN Architectures
Thesis · In ProgressIkramul Hasan Moral*, Md. Abu Bakar*, Samiur Rahman Omlan*, Fariha Islam*, Md. Touhidul Islam*
UIU Final Year Design Project · Supervisor: Dr. Muhammad Nomani Kabir
Which feedforward architecture best approximates a coupled nonlinear ODE system when trained purely on data? A controlled 69-run search over depth (1-4), width (20-100), five activations, and three optimizers, benchmarked against published PINN and DeepONet results on a solver-verified ground truth.
ANN Modeling of Hybrid Nanofluid Boundary Layer Flow
Research ProjectIkramul Hasan Moral*
Independent scientific ML project
A nine-layer neural network trained with Levenberg-Marquardt optimization to model hybrid nanofluid flow and heat transfer over a stretching sheet. Trained on about 32,400 physics-generated samples and validated against numerical solutions with MSE, RMSE, and R² metrics.
Talks & Teaching
Research Topics
Active Research Areas
Neural ODE Surrogates
Training the 69-experiment grid for the Lorenz-1960 thesis: separating architecture effects from optimizer effects, benchmarked against the PINN literature.
ML for Network Security
Extending the UNSW-NB15 baseline study with richer imbalance strategies, cost-sensitive learning, and cross-dataset generalization.
Research Data Collection
Built a live web instrument for collecting human perception data on memes for an ongoing research study.
Connect
Open to research collaborations, grad school conversations, and project discussions.