@@ -428,8 +428,6 @@ As our data is obviously not linearly separable we expected the __SVM Kernel__ a
The highest accuracy have the SVMs with features most important.This can be explained, among other things, by the fact that SVMs require few features. Too many features usually have a negative effect on the value of the machine prediction.
The results are also skewed by a most frequent bias. This is evident from the confusion matrix: for categories 2 to 5 (i.e., all categories except those for lowest alcohol consumption), there are almost no false positives. A large proportion of individuals actually report consuming very little alcohol (1); however, quite a few individuals are also incorrectly assigned to the lowest category.
Translated with www.DeepL.com/Translator (free version)
