What are target variables, predictor variables and prior probabilities?

What are target variables, predictor variables and prior probabilities? Are target variables like the target that you want to perform an algorithm on? Are predictor variables like things to predict the data example labels? Are prior probabilities supposed to be a variable or a number used to predict? Need help thanks!!!

Answer 1

[Shashank Prasanna]

Hi Lester, There are a lot of phrases here and I will try to explain them in that order with some context.

Target variable, in the machine learning context is the variable that is or should be the output. For example it could be binary 0 or 1 if you are classifying or it could be a continuous variable if you are doing a regression. In statistics you also refer to it as the response variable.

Predictor variables in the machine learning context the the input data or the variables that is mapped to the target variable through an empirical relation ship usually determined through the data. In statistics you you refer to them as predictors. Each set of predictors may be called as an observation.

Prior probability usually comes from the Bayesian Inference where you have prior belief that the probabilities of the parameters (or weights) come from a certain distribution.

You see that there are couple of different terminologies and this is because we have different branches fields like machine learning and statistics with different nomenclature. I would be able to clarify this further if you define what your training methods are and what your data looks like.

[Lester Lim]

Thanks for the reply!!! My training method is utilising LDA (Linear discriminant analysis) I am currently trying to implement it into my GUI program. My data are actually two pictures. The first is my training data and the second is my testing (picture I would like to apply on). I drew an ROI using impoly and label it. After labelling it, I apply the machine learning onto it. Throughtout this process, I've had trouble in defining the target, predictor and prior probabilities. Another issue I would like to clarrify is that are prior probabilities actually variables?

[Greg Heath]

The prior probability of class 1 in a mixture of classes is the expected value of the ratio of class1 objects to the total number of objects BEFORE any classification model is created. For a given class misture, the priors are fixed parameters and their sum is unity.

[Lester Lim]

So I have to calculate the expected value of the ratio of class1 objects to the total number of objects, from lets say I have 3 classes, so I will calculate and put all 3 probabilities into it. Thanks so much!!!

Answer 2

[David Willingham]

Hi,

Please see this page, it describes in detail target variables (predicted reponses), predictor variables and prior probabilities for supervised machine learning workflows:

Supervised Learning Workflow and Algorithms