Skip to main content
. 2020 Dec 17;10:22129. doi: 10.1038/s41598-020-79142-z

Figure 5.

Figure 5

Model variant comparison. The left panel demonstrates that employing the recurrent LSTM architecture after the model embedding layer leads to a significant increase in model performance. The right panel depicts the impact of differing the number of continuous value percentile bins. When considering 5, 10, and 20 bins, we find that 10 offers the best performance and use this in all other results. 95% Confidence intervals are bootstrapped from a 10-fold cross-validation using 10,000 resamples.