# Pattern Recognition within numerical data

I got different input data\instances and for each of them correspond different sequences of numerical data, which I normalized for comparison.

For example, instance1 has:

seq1: 1.3, 2.4, 1.0, 1.25 ...
seq2: 5.1, 3.9, 1.2, 7.8 ...


I normalized the sequences in different ways, for comparison. For example I got a normalization related with the starting value, or normalization intra values etc

I need to understand, to learn, pattern (if they exists) in these numerical sequences. For example, a trivial pattern would be: if my seq1 has a growing slope, it is likely that the next value will be greater than the previous one.

For this task I thought that a good approach would be the use of machine learning. For example using clustering, random forest, Decision Trees, or deep learning.

If I want to use algorithms like clustering, I need to specify the window on which I define my instance for the classifier, to learn the pattern. But in these way I cannot find pattern arbitrarily long.

What would you suggest? Is there a way to adapt solution like Convolutional NN to problem like this, to automatically extract pattern in the data (of arbitrarily long sequence)?

• The methods you mention are either keywords ('deep learning') or classification/regression methods (Random Forest, Decision Tree). Further you mention the concept of clustering. These are all targeting different problems/issues. If I understood your description correctly, you want to work on a problem of unsupervised learning since you have no target variable (such as in regression/classification). I recommend to read into the topic of unsupervised learning. Only then you can define your problem in detail and people in this forum might be able to help/assist. – Nikolas Rieble Jun 12 '17 at 12:23
• What is your objective? Do you want to forecast the next element in the series? To be clear, this is essentially a time series right? We need a little more info about the data. – Abraham Horowitz Jun 12 '17 at 12:41