Exploring household emission patterns and driving factors in Japan using machine learning methods

• Published in: Applied energy 2022-02, Vol.307, p.118251, Article 118251
• Main Authors: Chen, Peipei, Wu, Yi, Zhong, Honglin, Long, Yin, Meng, Jing
• Format: Article
• Language: English
• Subjects: Decarbonization; Factor analysis; Household carbon emission; Japan; Machine learning
• ISSN: 0306-2619; 1872-9118
• DOI: 10.1016/j.apenergy.2021.118251

•Multi-level household emission patterns are recognized by machine learning models.•Demographic structure, age and appliances are key factors to vary emission levels.•Customized emission reduction paths are proposed to guide energy-saving behaviours.•Household appliance usage obtains the largest reduction potential to mitigate emissions. Given by the ambitious GHG mitigation targets set by governments worldwide, household is playing an increasingly important role for reaching listed reduction goals. Consequently, a deep understanding of its emission patterns and the corresponding driving factors are of great importance for exploring the untapped potential of household. However, how to accurately capture household emission features still demand further support from both data and method development. To bridge this knowledge gap, we try to use machine learning technology, which is well linked to the micro-level household survey data, to identify key determinants that could explain the household home-energy consumption and associated emissions. Here, we investigate the household CO2 emissions based on a representative survey which covers 31,133 households in Japan. Six types of machine learning process are employed to find key factors determining to different household emission patterns. Results show that demographic structure, average age and electricity-intensive appliances (electric water heaters, electric heaters, etc.) are most significant driving factors that explain differences in household emissions. Results also further verified that differences in driving factors can be observed in identifying various household emission patterns. The results of study provide vital information for the customized decarbonization pathways for households, as well as discussing further energy-saving behaviours from data-oriented method.