Page 332 - Data Science Algorithms in a Week

Predictive Analytics for Thermal Coal Prices Using Neural Networks … 313

Developed Neural Networks to Predict Coal Price Relative to
The Future Quarter

The selected architecture and the selected set of inputs were utilized to establish a
final architecture. The neural network was trained with 63 samples. The next step is to
predict with 32 data samples of the 95 and neural networks with 8 and 12 input variables
(respectively) according to Sβ and the correlational method. The best result was obtained
with the neural network with 12 input variables (as illustrated in Figures 8 and 9).
Predicting the price of thermal coal was done with a lower error and capturing the
movements of the market, demonstrating the success of the learning ability of the neural
networks and the most important variables.

Figure 8. Prediction of thermal coal price relative to the future quarter using 12 input variables.

Figure 9. Prediction of the thermal coal price relative to the future quarter using 8 input variables.