57
Conclusions
nFor random behavior pattern, none of
the evaluated
methods is able to produce good results.
nFor stable behaviors, all algorithms
produce very good
results. More than 94% of the predictions have an absolute error less than
0.2.
nFor the trend, jumping, or two-phase
behavior patterns,
the BDES method outperforms the other three. Among its predictions, 93% to 99% have less than 0.2 absolute error. The
corresponding percentages
of the other methods are 10% to 40% lower depending the behavior patterns.