Hello!
I am Disha

ML Imputation & Explainable AI

My ongoing MSc dissertation — a large-scale experimental study on 1M+ records evaluating how imputation techniques affect predictive performance under varying missingness, training models with cross-validation and applying Explainable AI to compare accuracy, feature stability and fairness.

Road Traffic Accident Data Modelling

Designed a constellation-schema data warehouse on UK road-traffic accident data (2020–2024), built fact and dimension tables and multiple data marts in Hive, and developed interactive Tableau dashboards analysing accidents by severity, weather, vehicle type and demographics.

Retail Sales Performance & Demand Forecasting

Extracted and cleaned 500K+ retail transactions, applied statistical tools to identify demand patterns and seasonal trends, built an RFM-style segmentation model, and delivered a Power BI dashboard with 8 KPIs for real-time diagnostic analytics.

Wind Generation Forecasting Pipeline

• Built an end-to-end ML pipeline forecasting day-ahead wind power, using Germany as the target country on 35,000+ hours of real ENTSO-E generation data (2019–2022) pulled through Kaggle and merged with Open-Meteo weather reanalysis.
• Engineered 30 physics-informed features — cubed wind speed (Betz-law proxy), cyclical calendar encoding, and lag/rolling windows — then trained and compared XGBoost against Prophet on a leak-free chronological split.
• XGBoost cut error to 1,560 MW MAE (19.8% MAPE), far outperforming Prophet, and shipped an interactive Streamlit dashboard with four stakeholder views.