0
$\begingroup$

I have a database with different variables which contains information such as age, date of vaccination, number of doses as well as the number of antibodies (which is my target variable). If the number of antibodies is under 100 and 10, the patient's antibodies level is regarded as low and very low accordingly. My goal is to create a model by which I would be able to determine WHEN a patient should receive his next vaccination based on these facts. Thus far I have only generated decision trees using R studio. My mathematical knowledge is limited to come up with a solution. How could this be achieved?

$\endgroup$
2
  • 3
    $\begingroup$ If you're trying to predict dangerously low levels of antibodies, and you have continuous count data on antibodies, it would be better to model the continuous data as your outcome variable rather than the acceptable / low / very low binned version of that variable. (This is not binary BTW, so you would need ordinal logistic regression if you were to use this binned variable, despite the waste of information.) However, it's unclear to me whether you're really interested in simply predicting antibody count or whether you're trying to model longitudinal change, which is much more complex. $\endgroup$ – Nick Stauner Jul 2 '14 at 20:45
  • $\begingroup$ I think this question relates to supervised learning more than data mining (tags are inadequate). As for your question: asked that way, there are many ways of building a predictor. The right one to pick really depends on your data. Start with some reading on supervised learning: en.wikipedia.org/wiki/Supervised_learning ; if you don't know what to pick, you may try several ways and evaluate the performance of each method from a subset of your data (test database) not used in the learning process. Anyway, as @NickStauner suggests, predict the number of antibodies, not the binned version $\endgroup$ – Romain Reboulleau Oct 2 '18 at 15:50
-3
$\begingroup$

According to your question i can suggest you using logistic regression as you have two levels of dependent variable/Target Variable(DV) .Are your independent variable continuous or binary as well . You can also check similar post about binary DV (Prediction of a binary variable) and R tutorial on this (http://ww2.coastal.edu/kingw/statistics/R-tutorials/logistic.html)

$\endgroup$
2
  • 1
    $\begingroup$ I think this answer is wrong. The response variable is a count variable and dichotomizing it will loose information. You coud try with a usual linear model, maybe with a transformed response variable (such as log). But you should also make plots of the data, and diagnostic model plots. If you dont understand what this means, you need local statistician help. $\endgroup$ – kjetil b halvorsen Jul 2 '14 at 21:08
  • $\begingroup$ Adding to my comment above: log transformation of count data is usually not adviced, we have better methods nowadays. Try some Poisson (or quasi-Poisson) regression. $\endgroup$ – kjetil b halvorsen Oct 2 '18 at 14:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.