Skip to main content

Advertisement

Table 3 Household properties used to evaluate the household classification methodology; the relative class sizes in the complete sample and the finally used part, such as the mean and standard deviation of Accuracy and F1-scores for the Random Forest model based on all data sources are given

From: Mining volunteered geographic information for predictive energy data analytics

Property Class labels Class size Class size Accuracy (%) F1-Score (%)
   \(\mathcal {A}\) \(\mathcal {B}\) Mean (SD) Mean (SD)
Household type apartment 52.03% 69.23% 77.94 (5.30) 83.28 (3.01)
  house 47.97% 30.77%    66.36 (13.18)
Living area ≤ 95 34.07% 34.95% 59.78 (2.50) 65.00 (4.28)
  96−145 33.23% 40.61%    58.17 (4.10)
  > 145 32.70% 24.43%    56.40 (5.24)
Number of 1 person 13.07% 17.31% 58.08 (5.07) 50.39 (11.97)
Residents 2 persons 40.32% 43.01%    59.76 (5.03)
  3−5 persons 43.92% 38.12%    59.07 (6.09)
  > 5 persons 2.69% 1.47%    0 (0.00)
Heating type electric 12.86% 9.73% 90.05 (0.52) 68.08 (3.59)
  other 87.14% 90.27%    71.68 (4.30)
Water heating electric 50.49% 44.97% 70.03 (3.87) 68.08 (3.59)
type other 49.51% 55.03%    71.68 (4.30)