Figure 4 | Scientific Reports

Figure 4

From: Selecting the most important self-assessed features for predicting conversion to mild cognitive impairment with random forest and permutation-based methods

Figure 4

The figure shows feature importance using the ELI5 library62. The most important features are at the top and those at the bottom matter the least. The first column in each row shows how much model accuracy decreased with a random shuffling ± how the accuracy varied from one-reshuffling to the next. The most important feature is subjective cognitive decline (SCD), followed by diet features(sweets and white fish), hours of sleep during the day, and APOE. The rows in red show predictions of the shuffle data that happened to be more accurate than the real data. The idea behind this is that random chance caused the predictions of the noisy data to be more accurate, this indicates that the features do not contain information about the target feature.

Back to article page