Researchers have determined a new method based on machine learning that is said to reduce customer acquisition costs by approximately 15%, or $0.07/watt. It is based on a customized version of the XGBoost algorithm and takes into account, among other things, summer bills, household income and the age of the homeowner.
“We further dive into the modeling details of XGBoost and attribute its improved prediction performance in logistic regression compared to two factors: variable interaction and nonlinearity,” the researchers said. “Finally, we show XGBoost’s potential to lower customer acquisition costs and then the ability to identify new market opportunities for solar companies.”
According to them, this new method could help solar companies reduce customer acquisition costs and other soft costs associated with the residential solar business.
They compared the performance of the proposed algorithm with the logistic regression method, which the researchers described as the most commonly used method to analyze differences between PV adopters and non-adopters. “Our logistic regression model with nine original and highly visible household characteristics successfully predicts 71% of out-of-sample PVs,” ​​they explained. “The model correctly identified 66% of adopters and 75% of non-adopters.”
According to the research team, the modified algorithm was able to provide better results than logistic regression in predictive performance. “The predictive model correctly predicted 87 percent of the two PV adoption states, compared to 71 percent for logistic regression,” they added. “The correct adopter percentage increased from 66 percent to 87 percent and the correct non-adopter percentage from 75 to 88 percent.”
They attributed the superior performance of the machine learning-based approach to the fact that it integrates complex nonlinearity and variable interaction and takes into account factors such as summer bills, household income, and homeowner age.
“The advantage of using these variables is that they are easily accessible, so PV companies can collect data on them at low cost,” they also note. “Another reason for XGBoost’s improved performance is that it can potentially recover key latent information embedded in the data. “For example, including geographic information such as the respondent’s state or county increases the predictive accuracy of logistic regression to some extent.”
The research group estimates that the new method can help solar power companies reduce customer acquisition costs by about 15%, or $0.07/watt. It also explained that data mining and machine learning can also help reduce the soft costs of contract cancellations, supply chain management, labor allocation, and licensing and inspection issues.
It described the new methodology in a study published in “Machine learning reduces soft costs in residential solar power”. scientific reports. The research team consists of researchers from the US Department of Energy’s National Renewable Energy Laboratory (NREL), Lawrence Berkeley National Laboratory, Florida State University, University of Wisconsin-Madison and Renmin University of China.
