Once a leaf node is reached, a final decision on the food type is
returned. Our decision tree is created using the Matlab built-in
function classregtree, which applies Ginis diversity as
the separation criterion [34]. Multiple decision paths for a
single type is possible. This is because during the chewing
process of one bite, the sizes of the food chunks in the mouth
become smaller and smaller, which causes variations in the
feature values. However, we can still classify the chewing
events for the same food type by looking at different feature
combinations (i.e., different decision paths). When the food
chunks are small enough, in fact, it is almost impossible to
distinguish different foods. In such cases, since the major
acoustic features are so weak, the corresponding chewing
events are already ruled out in the event detection step.