The EnergyScore: An Alternative to Credit Requirements for Low-to-Moderate Income Community Solar
EnergyScore, a machine-learning algorithm, was created to improve access of community solar energy to lower income consumers. The aim is to decide which customers should be granted access in order to increase the number of customers while minimizing the risk of default. Currently the industry standard consists of a strict credit score cutoff which takes into account only one feature variable and the threshold of which has not been statistically optimized.
EnergyScore was trained with over 5000 financial and demographic parameters. We trained, validated and tested this algorithms on a dataset of nearly a million individuals tracked over 24 months. We used various classification and regression techniques in order to predict the probability that a potential consumer will complete their solar payments or be delinquent. We finally decided upon the most accurate model: a random forest algorithm, which consistently displayed accuracy rates at approximately 91%.
Authors: Sruthi Davuluri, Christopher Knittel, Chikara Onda
