Page 334 - Data Science Algorithms in a Week

Predictive Analytics for Thermal Coal Prices Using Neural Networks … 315

2
( ) = ∑ ( −ḟ ( )) (4)
=1
(1− ( ) ) 2

where CCV (λ) is the GCV for certain number of parameters (i.e., tree) as defined by
λ and the summation of the squared error is calculated for each training sample with
inputs xi and the desired output yi under the tree as defined by λ.
The training was conducted with 63 data samples for training and the most important
variables where the target was the future thermal carbon price. The following set of
equations represents the results of this analysis with regressions trees and the most
important variables which the coal price is modeled:

Y = 108.157 + 407.611 * BF6 + 367.188 * BF8 + 157.43 * BF9 70.7223 *

BF10 + 70.6882 * BF12 185.455 * BF13 (3)
where
BF6 = max⁡(0, 0-SUPPLY_COAL4);
BF8 = max⁡(0, 0.18-RENEWABLES_ENERGY_CHINA1);
BF9 = max⁡(0, CHINA_ECONOMY3 -6.84);
BF10 = max⁡(0, 6.84-CHINA_ECONOMY3);
BF12 = max⁡(0, 5.73-CHINA_ECONOMY2);
BF13 = max⁡(0, CHINA_ECONOMY3-6.64);

To verify the performance of the regression tree obtained with the 63 training
samples, the resulting equation was implemented in the 32 samples of testing data to
predict the price of thermal coal. Figure 10 shows the results.

Figure 10. Predicting thermal coal prices using regression trees.