# Recognize pattern in dataset

I'm currently working on a group project where we need to find a pattern in a given dataset. The dataset is a collection of X, Y, Z values of a gyroscope from someone who is walking. If you plot these values you'll get a result like this.

And this is how our dataset looks like.

We are new to AI and ML so we first did some general research like understanding how matrices work and how to do some basic predictions with frameworks like TensorFlow and PyTorch. Now we want to start on this problem. What we need to do is to find a pattern in the dataset and count how many times this pattern appears in this dataset.

We started of with some none AI functions to count, we managed to do that but the way we counted will probably only work on this specific dataset. So that's why we decided to do this with AI.

We would love to hear as many different approaches as possible to count the steps since we are still learning.

• Wouldn't a spectral analysis be more appropriate in this case? – firion May 22 '19 at 10:17

For such time-series data that has a significant amount of periodicity, I would recommend converting data to the frequency domain and performing various spectral analysis methods as @firion has already mentioned. For example, you could perform Fourier Analysis and study the individual components and identify patterns there.

Also, it generally not recommended to perform the normal pattern extraction approaches to time-series data as they fail to understand the temporal relationship between subsequent data points.

Hope this helps!

• since you are on ML forum, recognition of sequences does RNN.

• you wond belive, i currently work on similar stuff. Start about thinking of algo to find repititions in string, like : "ababcab" returns ('ab' : 0,2,5)

• and yes you could do Fourier Analysis but thats not ML method at all

You can also fit trigonometric functions like sin and cos. eg check this doc