About Me

School of Computer Science

Major: Evolutionary Computation and Intelligent Systems (ECIS) Lab

GPA: 3.88 / 4.30

Project: Development of a Web-Based Project Management Platform using Waterfall Technique

GPA: 3.49 / 4.00

(1) Researched interpretable regression modelling for interpolation and extrapolation using genetic programming (evolutionary optimisation), with a focus on quantifying and reducing predictive uncertainty/error as models are pushed beyond their training distribution, directly relevant to reliable risk prediction on unseen patients or data. (2) Designed and built an automated evaluation pipeline (Python, Git) systematically comparing five statistical model-selection criteria (MSE, AIC, BIC, MDL, PSM) across 20 benchmark datasets, following a rigorous, reproducible experimental methodology; found no single criterion is universally reliable, with effectiveness depending on dataset properties, work directly applicable to choosing among competing statistical/ML models in high-stakes prediction settings. (3) Developed hybrid model-selection criteria extending standard statistical metrics with three novel penalties targeting extrapolation divergence and interpolation/extrapolation sensitivity (a form of uncertainty-aware model selection), improving model reliability outside the training distribution. Validated on a real energy-system surrogate model, two resulting configurations (MVP-BIC, MVP-MDL) achieved ~98% improvement in test MSE over baseline without domain-specific tuning. (4) Built a synthetic data framework using KDE to identify sparse, high-uncertainty input regions, then used teacher models (NN, RF, GP) to generate augmented data and train GP student models; showed statistically significant improvements in extrapolation performance and reduced heterogeneous error across six benchmark datasets, a data-augmentation strategy transferable to limited-data biomedical prediction tasks. (5) Maintained reproducible, well-documented codebases and benchmark dataset repositories with automated test scripts, leading to peer-reviewed publications at EuroGP, GECCO, LOD, and ECTA.