ZEMCH 2019 International Conference Proceedings April.2020 | Page 23
3.2. DT analysis results
Figure 1. Decision tree model for EUI for heating (Nodes with bold outline are terminal nodes, and mean of
each terminal node represents the prediction. Separation point is set to 0.05)
3.2.1. Determinants of EUI for heating
The decision tree result for the EUI for heating is given in Figure 1. As described in Section 2, DT
model can identify nonlinear relationships among variables that cannot be found in MLR model.
Accordingly, the first separation occurred by the year of building permit which was not found in MLR
model. The average EUI for heating of the latest building permit group (node 3), is much lower than
that of the middle group (node 2). This result is highly acceptable since all samples in the latest group
are permitted after 2008, in which the South Korean design standard for the U‐value of apartment
building envelope has been drastically strengthened. However, unexpectedly, the average EUI of the
oldest group (node 1) is lower than that of the middle group. One possible reason for this result is
uneven distribution of samples. As shown in the 3 rd row in Figure 1, the distribution ratio of samples
with heating set temperature over 26.0℃ is higher in the middle group than in the oldest group,
leading to higher average EUI.
Heating set temperature is the second major determinant. Units with set temperature over 26.0℃
use more energy for heating than others. In addition, surface area and use of auxiliary heating devices
are also found as significant factors in specific sample groups.
Analyzing Determinants of Energy Consumption for Heating and Cooling in Apartment Units –
Comparison of Linear and Nonlinear Statistical Models
12