Projects
Two open-source research-engineering projects
Projects
Beyond the formal publications, two open-source research-engineering pipelines support the climate-disease modelling work. Both prioritise reproducibility, deterministic runs under a fixed random seed, and a clean separation between data, feature engineering, model fitting, and reporting. Each is the engineering deliverable for one of the formal publications.
TyphoidNepalML — Research Pipeline
End-to-end, deterministic Python pipeline implementing the three-model framework (Random Forest · XGBoost · MLP) for the research Manuscript. Reads four national-scale CSV inputs (HMIS surveillance, CHIRPS precipitation, ERA5-Land temperature and humidity, DRR-portal flood records), performs district-name harmonisation and Bikram-Sambat → Gregorian calendar conversion, engineers lag features and rolling means, fits the three architectures under a strictly chronological 12-month held-out test split, and emits a four-figure publication-quality figure set plus a structured analysis report — all in roughly two minutes on CPU.
| Run | cd manuscript && python3 main.py |
| Architecture | src/loader.py · src/processor.py · src/engineer.py · src/models.py · src/evaluator.py · src/visualizer.py · src/reporter.py |
| Driver | manuscript/main.py |
| Outputs | manuscript/output/figures/ · manuscript/output/analysis/research_analysis.txt · manuscript/output/tables/performance_metrics.csv |
| Reproducibility | Pinned dependencies in manuscript/requirements.txt · fixed random seed (42) · committed manuscript/data/ |
| Publication | Research Manuscript |
| Repository | GitHub source — manuscript/ |
TyphoidNepalML — Thesis Pipeline
End-to-end, deterministic Python pipeline implementing the four-model framework (Random Forest · XGBoost · MLP · Weighted Ensemble) for the Master’s Thesis. Extends the research-pipeline framework with an SLSQP-weighted ensemble layer that achieves R² = 0.8675 on the held-out 12-month test window. Emits a five-figure publication-quality figure set (national time series, correlation heatmap, XGBoost feature importance, four-model actual-vs-predicted, seasonal boxplot), four trained model artefacts (RF, XGBoost, MLP, and scalers), three CSV tables (descriptive statistics, lag correlation, performance metrics), and a structured academic-analysis report.
| Run | cd thesis && python3 main.py |
| Architecture | src/loader.py · src/processor.py · src/engineer.py · src/models.py · src/evaluator.py · src/visualizer.py · src/reporter.py |
| Driver | thesis/main.py |
| Outputs | thesis/output/figures/ · thesis/output/tables/ · thesis/output/analysis/research_analysis.txt · thesis/output/models/ |
| Reproducibility | Pinned dependencies in thesis/requirements.txt · fixed random seed (42) · committed thesis/data/ |
| Publication | Master’s Thesis |
| Repository | GitHub source — thesis/ |
Both pipelines are open to academic and public-health collaboration. For pilot deployment or partnership inquiries, email.