Nested classifications

Question

I was wondering if anyone can point me to anywhere or in the right direction. Which approach is the best for tackling a nested classification problem.

So we want to classify species of animal say etc. We have split this into three layers - cellular level, organ level, organism level. So we have different real world tests (or sources of input data from animal fares etc from around the world) - each of these provide data at a cellular level, and organ level and the organism level.

So a dog can be classified by its outward appearance, or features based on its internal anatomy, or indeed its DNA etc. And knowing something about the DNA can tell you something about organs, and organs can tell you something about the overall structure.

So if I wanted a classification based on the cellular data, to be used to rule out or help classify the next level, and then that to feed into the top level - which approach is best served for this problem? Is this a feed forward neural network type of problem, hierarchical classification? A deep learning aspect?

This problem can apply to cars, at the engineering level, then at the consumer level etc..... so based on its parts, you can tell which type of car it is etc. So knowing its parts, and some features of its outward structure you are more confident in your classification.

Answer 1

In my opinion, deep learning algorithms are one of the algorithms that you could use. You really don't have to nest the classification but instead use all of the data or choose the features that you think would be useful.

Convolutional neural networks could also be used. refer here. or here. Although this is mostly used in computer vision, and image classification but with you're problem you probably might want to apply this where your "layers" are to be determined by your model and you'll never know, it might find more relationship amongst your features than you think.

Answer 2

On Hierarchical Classification I think you should check this survey article