Sam Hardy | Monday, September 9th, 2019
Problem: Historic sales data not driving NHS purchasing decisions.
-Collaborated with stakeholders to agree objectives, target features, data inputs and expected improvements.
-Collaboratively agreed success measures.
-Agreed a ‘hold-out validation set’ evaluation protocol with stakeholders.
-Extracted, transformed, loaded data from multiple disparate NHS databases; based on agreed data extraction policies including: missing data, categorical (ordinal/nominal) data and feature engineering.
-Trained baseline model leveraging ‘training’ data.
-Tuned model/algorithm (reducing overfitting/underfitting) using cross-validation techniques on ‘validation’ data.
-Tested model on ‘test’ data; using Power-BI.
Outcome: NHS Purchasing able to identify sales opportunities and predict pricing changes; improving purchasing decisions.