Optimization of In-House Energy Demand - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 277

Table 10.2 Cross-validation of trained models
NB FHMM CT 1NN
(a) Classification accuracy of device states per household averaged over all appliances
House 1 0.8429 0.8414 0.8319 0.7615
House 2 0.9310 0.9300 0.9224 0.8062
House 3 0.9275 0.9200 0.8908 0.7213
House 4 0.8645 0.8616 0.8746 0.7038
House 5 0.9864 0.9854 0.9839 0.7638
House 6 0.8131 0.7873 0.7752 0.6050
MEAN 0.8942 0.8876 0.8798 0.7269
(b) Classification accuracy of device states per appliance averaged over all households
Air conditioning 0.9315 0.9248 0.9300 0.9138

Bathroom GFI 0.9328 0.9275 0.9324 0.9134

Dishwasher 0.9541 0.9493 0.9551 0.9134

Disposal 0.9955 0.9818 0.9970 0.9918

Electrical heating 0.8863 0.8856 0.8620 0.8895

Electronics 0.8875 0.7970 0.7404 0.0991

Furnace 0.8216 0.8211 0.7294 0.5216

Kitchen outlets 0.7902 0.7915 0.7070 0.1775

Lighting 0.7751 0.7737 0.8006 0.7611

Microwave 0.9516 0.9473 0.9526 0.9279

Miscellaneous 0.9242 0.9295 0.9296 0.7237

Outdoor outlets 0.9982 0.9995 0.9997 0.9996

Oven 0.9754 0.9804 0.9815 0.9811

Refrigerator 0.7834 0.7872 0.7952 0.7898

Smoke alarm 0.9729 0.9629 0.9738 0.6234

Stove 0.9346 0.9288 0.9363 0.8330

Subpanel 0.9808 0.9807 0.9815 0.9811

Unknown outlets 0.9578 0.9558 0.9555 0.3432

Washer/Dryer 0.9287 0.9256 0.9297 0.8763

MEAN 0.9148 0.9079 0.8994 0.7505

(1NN) classifier. The classification results present the performance of the trained
models in an unknown environment or rather before unseen household.
The results in Table 10.2 a show the classification accuracy of device states per
household averaged over all appliances. For instance, the NB model achieved an
accuracy of 0.8429 for House 1, meaning that the model was trained on House 2-6
and tested on the previously unknown House 1, where 84.29% of all device states
were classified correctly. However, as illustrated in Table 10.2 a the classification
accuracy of each model varies with the household, which is due to the fact that
the examined households use appliances of different manufacturers with dissimilar
energy profiles.