Figure 13.

Figure showing the top 29 features by importance of an example random forest model using the Gini index. The prefix refers to a particular gradient/measure across the electrode grid. The suffix refers to a single electrode feature.The feature ‘start’ can be manipulated to calculate the wavefront direction. As the random forest model is trained, the reduction of the Gini index caused from each feature is measured and averaged across the whole decision tree ensemble. From this particular random forest, the mean maximum voltage and standard deviation of the start index are important features. It can also be seen that from the magnitude of the Gini values, no individual feature is suitable just by itself. These plots help determine suitable feature sets for future models.