Week 5 – Transitioning to Machine Learning

24 Jul 2026 - Sneha Dharmeche - climate risk, data science, feature engineering, NHS ERIC

Project overview

This week was the transition from data preparation into predictive modelling. I completed the main data cleaning stage and finalised the dataset so it could be used for machine learning.

The focus for this week was final data cleaning changes. I created an age vulnerability score to represent the proportion of older buildings within each NHS site, allowing building age to be incorporated as a single predictor.

I also created geographic coordinates (latitude and longitude) from site postcodes using pgeocode, enabling spatial analysis and the integration of location-based information.

Another important task was handling missing data. Rather than removing sites with incomplete information, I created missingness indicator columns for features containing missing values and applied median imputation to estimate the missing numerical values. This method retains all sites within the dataset while allowing the model to distinguish between observed and imputed values.

With the dataset prepared, I began building the first predictive model using Gradient Boosting to investigate which estate and climate-related factors are most strongly associated with NHS costs.

The coming weeks will focus on refining the model, evaluating its performance and comparing its results with a Bayesian modelling approach.