I am kind of new to the data mining subject but i need help to choose a learning algorithm for my application:
The problem: identifying that a certain curve or data set belongs to a certain fault in a Component.
My training data should be like this:
Motor Current Values:
[0.5,0.6,...,0.4] -> Fault in Komponent 1
[0.2,0.3,...,0.4] -> Fault in Komponent 3
[1,0.7,...,0.4] -> Fault in Komponent 2
[0.8,0.7,...,0.3] -> Fault in Komponent 3
And i was wondering if i should using a cluster analysis (k-means and save centroid centers then compute the distance of each new entry then give a fuzzy estimation to where which cluster my data might belong to).
Decision Tree algorithm(entropy of the values and stuff), Or should i calculate a
distance between the Nominal data (Healthy(No fault) Curve) and the faulty
data and play on a more simple basic decision tree with thresholds ?
Proceed with a peak analysis and count the number of peaks within my data and add it to my learning algorithm.
And when should i do a data pre-processing like normalization ?
This is an example of my entry parameters :
[Speed(mm/s), Motor Current Values(Ampere), DistanceToNominalState(Ampere)]
Any suggestions ?