Gradient Boosting Approach to Predict Zero Carbon Achievement of Households in Kitakyushu

The objective of this paper is to develop a gradient-boosting algorithm based machine-learning model that predicts the zero-carbon achievement of households using 11 months of energy data and questionnaire survey results, obtained from 198 smart houses located in Kitakyushu, Japan. In order to predict the zero-carbon achievement of households, we use LightGBM (light gradient boosting machine) classifier for feature selection and demonstrate that the most important features are the mean absolute error (MAE) of housing type, floor heating, average monthly electricity bill, and average monthly gas bill. This light GBM-based model can also predict the zero carbon achievement of households that did not participate in the questionnaire survey. This model helps us determine households with low carbon emissions and factors that influence zero carbon achievement. The work is currently ongoing, and the results presented in this paper are encouraging and provide valuable insights for developing effective household energy policies, which is a big step towards achieving nearly zero-carbon houses.