1

The main challenge is the lack of data. Input values come from tests results of patients. A patient takes a breath test at an interval during a timespan. The result values can range from 0 to ~200, and can be plotted for diagnose by a doctor based on the curve shape. I am looking for an algorithm that takes the values at every interval and comes up with a single output value from 0 to 1 that indicates a fuzzy boolean to determine a preliminary diagnosis.

As mentioned the data is scarce, the current data available is a set of around 10 typical values of positive and negative test results.

I am looking for an algorithm or model that can rely on the available data.

I looked into Fuzzy Logic which could be an option despite I am not convinced.

What are your thoughts?

human
  • 111
  • 1
  • Welcome to DataScienceSE. Afaik fuzzy logic is not relevant here. What you need is a supervised ML method, either classification or regression. Most classification methods can output a predicted probability, that would correspond to the "fuzzy boolean" output. When you say that you have 10 typical values, do you mean 10 sequences of 200 values or 10 individual values? The former case is quite small for a training set, in the latter case you can't even train a sequence model. Of course it might be simple enough that you just need a heuristic: did you look whether the mean is a good indicator? – Erwan Feb 28 '21 at 16:48
  • @Erwan the sample data corresponds to about 20 patients (corrected from 10). Each patient underwent a breath test every 15 or 30 minutes for 3 hours. Each patient in each session had about 12 breath tests, each data set consists of 20 rows and up to 12 columns. A mean would do no good since peak times are more informative than overall values. I believe a heuristic approach may be challenging. – human Feb 28 '21 at 20:20
  • You could try to train a sequence classification model. I don't know much about this myself but I suspect that your training set might be too small. – Erwan Mar 01 '21 at 00:35

0 Answers0