Hi VisteonIntern! Thanks for your critical comment and your interest on my app! I will try to explain you my
thoughts :-) The first thing is i didn't use k-means
which is a clustering technique which clusters a set of
datapoints in a predefinded number of clusters and you
are absolutely right that it is hard to determine the
number of clusters before. Instead I used k-nearest
neighbour classifier (I admit that the two names can
esily be confound). Since I don't know if you are
familiar with the concept of classifiers I will give you
a short introduction: A classifier works as follows:
1. Step: trainig phase - in the training phase let me say you have to create the "world" with its elements the classifier should know. Let me give you an example: If you want to classify music acording to genre and you have four genres (classical, jazz, rock, rap), then you
have to tell the classifier what are the characteristics
of the different genres. So you will take many music
clips (samples), calculate some features (beats, onsets,
pitch, timbre ..) and train the classifier
2. Step: classification phase: now you will show the classifier a new music piece (sample) which he has never seen before. This peace will be compared with the peaces that were used in the training phase. According to the operation mode the classifier will then find the genre class to which the new peace belongs by mathematical
calculations. In case of k-nearest neighbour the euclidean distance from the new sample to all samples
from the training phase is calculated. According to the
choosen number of "K" the k nearest samples from the
training phase to the new sample are considered. If the new sample is mostly e.g. near class "classical" the
classifier wil return class classical as result for the new sample. ---- now to your approach to use fuzzy logic. I think fuzzy logic is also a classification technique, but I must admit that I am not so familiar with it but I know that fuzzy logic is mostly used when there is no mathematical description of the "world" but a verbally like in speech recognition or automation of technical processes. -- I hope I could clarifiy your doubts a little bit ;-) if not feel free to ask me again or to tell me what you think is wrong with my approach!
User2858