This is a the final product of Machine Learning in Python coursework. Alongside my colleague A, we examined the effects of The London Energy Efficiency Pilot on energy usage. The project allowed us to support one another through a shared dataset, but work independently to answer different research questions.
I proposed the following question: Can we predict the dynamic pricing tariff class label (high/normal/low) based on energy usage patterns, ACORN class (a measurement of affluence), half-hour increment, and month?
Question Context: Having clear delineations of usage per ACORN group, per month, per half-hour would not only help to determine the effectiveness of a pilot, but also allow thresholds to be established in the broader sense of when price tariffs might be most effective in behavioural modification. The ideal model would allow decision makers to explore the cusp of mean energy usage under each tariff label and test the thresholds that cause a consumer to stratify their energy usage.
The pilot uses cost as a deterrent for energy usage during tariffs, but if tariff labels are unknowable by machine learning models, this might otherwise suggest that the energy usage is determined much more by external factors (such as weather, household wealth, culture, and many more) than the pilot. For example, a household might ignore price tariffs to keep their house warm on a cold day and simply endure the extra cost.
Data: Data comes from Smart Meters of London and is too large to be stored on GitHub.
Code is here.
Analysis, Visualization, and Model Scores are here.