ZEMCH 2015 - International Conference Proceedings | Page 329

building systems scored better than those with stand-alone systems.
• Buildings that made use of Facilities Management systems, building commissioning and energy & economic modelling scored better.
• Case studies with an intelligent HVAC control system scored better than those with intelligent
lighting systems.
• IBTs that laid more emphasis on energy and cost efficiency scored better than those which
emphasised comfort and convenience; network connectivity and AV control; and water saving
features.

Relationship Development
4.1 Regression Analysis and 5-Fold Cross Validation
The BREEAM and LEED certified buildings were separately, randomly partitioned into 5 equal size
subsamples but care was taken such that each set contained a mix of all the rating categories. At
every fold: of the 5 subsamples, a single subsample was retained as the validation data for testing
the model, and the remaining 4 subsamples were used as training data. The cross-validation process was then repeated 5 times, with each of the 5 subsamples used exactly once as the validation
data. The training sets helped obtain the best-fit model and the corresponding test set helped
validate the accuracy of that model.
The models were developed using percentage BREEAM/LEED scores as the dependant variables
and the number of IBTs as the independent variables. The regression models obtained at every
fold (linear, logarithmic, quadratic and cubic) were tested on the corresponding test sets so as to
obtain an insight on how the models would generalise to an unknown data set. This procedure
estimated how accurately the predictive model would perform in practice i.e. accurately predict
the BREEAM/LEED scores of buildings depending on the IBTs used. Also the standardised errors
of each of the models were calculated by using the Predicted versus Actual BREEAM/LEED scores,
and reported at every fold so as to identify the best-fit predictive model. Also the R and R2 values
of each model were reported and an Anova and T-test of significance were conducted.
After reviewing the R and R2 values and the average standardised error of all the models across
every test set, the logarithmic model was considered as the best fit predictive model for both
LEED and BREEAM. It was observed that though the R and R2 values were significantly high and
positive in value in all cases, suggesting a strong positive correlation between the two variables,
the logarithmic model performed very highly in comparison with the other models.
To describe the behaviour of the logarithmic model in detail a test set of BREEAM certified buildings has been displayed below (Figure 6):
• All of the buildings that have used 0 to 5 numbers of IBTs in them and have scored either a
GOOD or a VERY GOOD rating have caused the graph to rise exponentially in almost a linear
manner at the start.
• The buildings in the curved portion of the graph have used 6 to 9 numbers of IBTs in them and
scored an EXCELLENT rating. It can be observed that in these buildings the relative rise in the
BREEAM score with the number of technologies used is not as drastic as observed in the earlier
set of buildings causing the linear graph to slowly start to curve.
• Buildings G1-1, 3, 11, 12 and 21 follow a stabilising rise along the curvilinear path. All of these
buildings have used 10 to 16 numbers of IBTs in them and only scored an EXCELLENT rating. It
can be observed that though these buildings have a higher number of IBTs used in them their
BREEAM scores have not drastically increased, as has occurred with the buildings with lower

Building intelligence and sustainability using leed and breeam in the UK and Europe

327