Predict probability of a customer cancelling their utility contract

Data science demo by Thomas Wood, Fast Data Science Ltd

This is a machine learning model that predicts the probability of a customer cancelling their contract with an energy company. Normally a company of this size would know lots of information about its customers such as their usage and spend patterns. However it's still difficult to manually predict exactly how likely a customer is to cancel their contract and go to another provider, and this is why machine learning is useful.

The model is built with Random Forest with 10,000 estimators.

You can try out the model by adjusting the parameters and watching the effect on the calculated probability. Note that in some contexts, changing a value may not affect the probability. Also probabilities are always quite low since the majority of customers do not switch providers in a 3 month period.

electric consumption over last year
gas consumption over last year
electric consumption over last month
the forecast baseline electricity use over the next month
the forecast baseline electricity use over the next year
the forecast electricity use over the next 12 months
the forecast electricity consumption over the next month
the forecast electricity consumption over the 12 months
the forecast electricity use over the next year
the forecast value of discount on contract
the forecast standing charge for the next 12 months
the forecast energy price over the next 6 months
the forecast energy price over the following 6 months
the forecast power price over the next 6 months
does the customer have gas
the current usage of the customer
the gross profit margin of the company
the net profit margin of the company
the number of products the customer has bought
the total net margin of the company
the number of years the customer has had a contract with the provider

Churn probability