Let's say I have 10,000 training points, 100,000,000 points to impute, and 5-10 prediction variables/parameters, all numeric (for now). The target variable is numeric, skewed normal with outliers. I want to use SVM, but I'm new, so I would appreciate any opinions.
6
-
$\begingroup$ Welcome to DataScienceSE. The question is not clear, please give more detail. I'm not sure that you use the word "impute" with its usual meaning: imputing is replacing a missing value with some replacement value obtained from the other instances. Do you mean that you want to predict the variable as target? $\endgroup$– ErwanCommented Mar 31, 2022 at 12:30
-
$\begingroup$ Thank you! Maybe prediction is a better word, but we are indeed trying to fill in blanks in our data using values obtained from non-blanks, based on their orientation/proximity in parameter space. I am looking into techniques like kNN and SVM right now. I am, going to try this either way so I am just curious about people's opinions on this type of application. What other detail should I have provided? I tried to describe the data without getting into the specifics. $\endgroup$– StonecatCommented Mar 31, 2022 at 12:51
-
$\begingroup$ Ok, from a ML point of view you're doing a regression task (as opposed to classification where the target is categorical). Usually SVM is used for classification but there's also the corresponding method called SVR for regression. Imho it's certainly one of the good options available, but of course it depends a lot on your data. Don't forget to keep some of the data where the target is know for evaluating the model. $\endgroup$– ErwanCommented Mar 31, 2022 at 15:57
-
$\begingroup$ Wonderful, thank you so much for your help with the terminology. Can you think of any other ML techniques that you would recommend for this application? $\endgroup$– StonecatCommented Mar 31, 2022 at 18:36
-
$\begingroup$ There are many options: linear or logistic regression, decision tree regression, k-NN... and I'm probably missing a few others. $\endgroup$– ErwanCommented Mar 31, 2022 at 19:28
|
Show 1 more comment