Information Technology Reference
In-Depth Information
Figure 16. Complexity optimization CV error on tone cartridge man. dataset
Complexity optimization CV error on tone cartridge
manufacturer's dataset
86000
76000
66000
56000
46000
36000
26000
16000
Complexity
Figure 17. Complexity optimization CV error on Statistics Canada dataset
Complexity optimization CV error on Statistics Canada dataset
880
860
840
820
800
780
760
Complexity
ing the ANN's performance, it turns out that even though its performance was strong on the chocolate
manufacturer's dataset (Table 1 - Rank 4), it had extremely poor performance on both the toner cartridge
manufacturer dataset (Table 2 - Rank 19) and on the Statistics Canada manufacturing dataset (Table
3 - Rank 16). As a result, the Super Wide artificial neural networks with cross-validation based early
stopping method was disregarded as a potential alternative.
The Super Wide multiple linear regression (MLR) was much closer with an average error across the
two manufacturer's dataset of 0.7353 compared to 0.7516 (significance 0.0274) for the automatic expo-
nential smoothing (ES) and compared to 0.7154 (significance 0.0000) for the Super Wide SVM. The
Search WWH ::




Custom Search