Hypothesis-driven research identifying novel concepts and proving them with rigorous methodology.
Architecting a novel, modality-aware Transformer framework to diagnose neurodegenerative diseases from multi-modal neuroimaging data. This work pioneers a 'modality dropout' scheme to build a system robust to the real-world challenge of incomplete clinical data.
Led a team research project developing a novel Knowledge Graph Separation methodology for quantifying structural alignment of LLM embeddings with biomedical knowledge graphs following QLoRA fine-tuning. Demonstrated 126% improvement in embedding geometric alignment with Phi-3 Mini.
Comprehensive ML pipeline analyzing 5,600+ exoplanets from the PHL Exoplanet Catalog to predict habitability. Implemented multiple classification algorithms with model explainability using LIME to interpret planetary characteristics influencing habitability predictions.
Research study investigating the impact of gamified features on mobile banking app adoption and user engagement among university students in Bangladesh. Comprehensive statistical analysis of 49 respondents with all four hypotheses achieving statistical significance (p < 0.05).
Predictive model analyzing 114,002 Spotify tracks to forecast song virality. Used clustering-based virality definition with K-Means and multiple classification algorithms achieving 100% accuracy with regularized models.
Developed a stacking ensemble model that achieved 96% accuracy and 100% recall in forecasting U.S. recessions. Used SHAP to interpret the model and validate its economic logic.