Nearest neighbor method?

Question

Machine learning

This method can be used to classify by distance, weight, etc. But for the classifications of what it is really used in real life?

Answer 1

What does the classification "by weight, by distance" have to do with it?
There is a classification problem - i.e. assigning an object to one of predetermined classes. There are many "families" of classification algorithms, such as tree-based, but SVM-based, and so on. and including a group of methods based on the metric representation of the space of description of objects of classification. The nearest neighbor method you mentioned is just one of the last of the group.
The methods of this group work according to the following scheme. Objects are represented as vectors in the N-dimensional feature space, and in this space some kind of proximity/distance meta must be introduced (which, by the way, is not possible in every space). There can be many of these measures: the well-known and familiar Euclid measure, Manhattan distance, Hamming distance measure, Chebyshev distance, Minkowski distance, non-metric cosine distance, etc. The choice of a specific measure is, generally speaking, a non-trivial task, the solution of which depends on the scales of these description parameters, the semantics of the task, and even on the specific classification method that will be applied.
Therefore, the answer to your question may sound like this - this method can be applied wherever an applied problem can be embedded in the scheme of the metric classification method. An example that probably everyone has heard is the solution of the problem of diagnosing oncological diseases in the IBM Watson system. Another example is intrusion detection in IDS systems. Predicting the creditworthiness of borrowers is an example that is also on everyone's lips. Identification of mineral deposits. I almost forgot - a classic of the genre - spam detection.
But in fact, hundreds of practical problems that are solved in this way are described.