Clustering Problem - Senseless output

Hey everyone,

Ich am working with a 1440x50 data set --> so i have 50 examples with 1440 timesteps. When I cluster it wit the Neural net clustering app, I always get one cluster, which contains many samples and then a lot of lone examples in different clusters.

I rather expected, to get like 4 samples in cluster 1, in the other one 5, in the other one 2.. and so on.

When I do 10 x 10 Layers on 50 examples... i have on cluster with 14 examples... and all others are spreaded around lonely. When I cluster the "lonely" ones again, I have the same behaviour. That the net clusters 7 samples together and the other ones lonely again.

Can somebody help me.. that I can get with one Clustering net ... more and better filled clusters?

Answer 1

[Yogesh Khurana]

nctool is the Neural Network classification or clustering tool that leads you to solving a clustering problem. The majority part in getting a clustering output is played by the data provided to the tool. In case of using nctool you are training a neural network. So, the number of neurons decided will also play some role. Also, setting some of the parameters while moving forward in the nctool app will help you in getting good clusters (if your data is for what you are expecting the clusters).

I am providing you some links to refer in order to use all the parameters of nctool in a productive way. Please refer to the following links:

https://in.mathworks.com/help/deeplearning/gs/cluster-data-with-a-self-organizing-map.html

https://in.mathworks.com/products/deep-learning.html#net

https://in.mathworks.com/help/deeplearning/ref/neuralnetclustering-app.html

Hope this helps!