Machine Learning: Classify Array of Numbers based on Patterns


I have experimented with various regression/classifier libraries, they accept training input like "5 is bad" and "10 is good" so they can tell you 7 is bad and 8 is good.

Now imagine a more complicated example: I have heart rate data and there are certain heart rate patterns that are good and other patterns that are bad. In this case it's nonsense to just say "heart rate 120" is good or bad and so forth. How do you tackle a problem like this? Are there any known ML/AI algorithms that can intelligently recognize patterns?

A good start could be a classifier that I can at least train with a whole array of numbers instead of a single number?


Posted 2018-07-10T12:52:36.373

Reputation: 3

RNNs like LSTM handle sequential data well – Alex – 2018-07-10T15:31:40.417



Actually you want to classify time series. Going from here you have basically two options:

  1. Build features from this time series, i.e. RMS, Peak Values, etc.. and classify them with "classical" predictors like SVM or Random Forests.

  2. Use neural networks, or more specifically use LSTM in order to feed directly your "array of numbers", which is a time series.

Andreas Look

Posted 2018-07-10T12:52:36.373

Reputation: 863