Modeling air conditioning ownership and availability

In recent years, extreme heat amplified the need for indoor cooling systems. Some local governments started to provide household cooling systems for vulnerable people. However, the lack of local AC prevalence makes equitable AC dissemination more difficult. Therefore, this study developed neighborhood-level AC prevalence information for California (CA), US Properties were on the market from 2018 to 2021 with a unique real estate dataset from Zillow Transaction and Assessment Dataset (ZTRAX). The analysis selected housing attributes with theoretical or empirical linkages to AC access. Data preprocessing imputed missing values with a random forest (RF) analysis. Next, a subsequent multi-class RF estimated the types of household AC ownership (central, other, yes, and none). The RF model showed an overall accuracy of 98% (class-specific accuracies for Central: 98%, None: 96%, Other: 95%, Yes: 99%). Jackknifing revealed the latitude, longitude, heating type or system, elevation, built year, cooling degree days, and building quality exhibited the highest Gini importance values. This study visualized AC prevalence in CA and four counties with major cities: Los Angeles, San Diego, Sacramento, and San Francisco County. This result can be applied to implementing heat prevention measures such as providing household cooling systems, energy subsidies, cooling centers, and increased green space access.

Automated summary

This study developed neighborhood-level AC prevalence information for California and predicted household AC ownership types using a random forest model. The results identified the key factors influencing AC prevalence, such as latitude, longitude, heating type or system, elevation, built year, cooling degree days, and building quality. These findings can inform heat prevention measures, including providing household cooling systems, energy subsidies, cooling centers, and increasing green space access.