1
$\begingroup$

I have a table full of different applications which have various fields that I want to use as features for my model (i.e. vulnerabilities, assessment results, etc.). These applications also have a field of the number of incidents that have occurred within them in the past year.

I am trying to find the best ML model to predict the probability of an application having an incident. I intend to train this model on the applications which have already had incidents, where the more incidents an application has then the stronger the weights should be on that application's features. Then with the resulting model, I want to test the other applications that have not yet had an incident to determine the probability that they will have an incident based on their features.

For the machine learning classifiers that I know of (i.e. logistic regression, neural networks, bayesian classifiers, etc.), I am only familiar with these giving simple 1 or 0 classifications of some input falling into a particular class. I am instead interested in a concrete probability of an application falling into a particular class, which is the probability they will have an incident.

$\endgroup$

1 Answer 1

1
$\begingroup$

For the machine learning classifiers that I know of (i.e. logistic regression, neural networks, bayesian classifiers, etc.), I am only familiar with these giving simple 1 or 0 classifications of some input falling into a particular class.

This is not true. Most machine learning algorithms make predictions in some kind of score, that can be used for making hard classifications (0 or 1). The score is usually bounded between zero and one and can be interpreted as a probability. The algorithms you mentioned are examples of such models that predict probabilities.

One thing you may take into consideration, is that some algorithms may return probabilities that are not well calibrated, so may need additional calibration of the probabilities.

For a start, you could try logistic regression, simple, but probably the most popular algorithm for such purpose.

$\endgroup$
1
  • 1
    $\begingroup$ In particular, logistic regressions and neural networks (with appropriate activations on the output, like softmax) give probability outputs. Converting to a binary category depends on some threshold that can be tweaked as the user deems appropriate and does not have to be $0.5$. $\endgroup$
    – Dave
    Jul 6, 2020 at 15:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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