6

I think a lot depends on what the purpose of the model is. Inference or prediction ? If it is inference then you really need to incorporate some domain knowledge into the process, otherwise you risk identifying completely spurious associations, where an interaction may appear to be meaningful but in reality is either an artifact of the sample, or is masking ...


5

No. If you find a correlation between a continuous variable and some event has some value r, the correlation between that variable and the negation of that event is not 1-r, but -r. If the spread percentage has a correlation of 0.17 with the team winning, it has a correlation of -0.17 with the team losing. Think of it this way - the positive correlation of ...


3

Yes, elastic net is a form of data dredging. It represents safer data dredging because during the selection process it carries along with it a penalty that "remembers" the context. Like lasso and ridge regression, a parameter will have a larger penalty when there are more candidate predictors. So the coefficient for selected variables will be ...


3

If I understand correctly, the shape (5244,19) corresponds to 5244 windows with dimension 19 right? If that so, I think that for solving your problem you should expand your X with one more dimension obtaining a shape of (5244, 19, 1) meaning (windows, window_size, number_of_sensors). TSFEL is flexible enough to extract features from multiple sensors at the ...


2

You don't need a boosted tree, or even interaction terms/deep trees, to get this type of behavior. This is an example of omitted-variable bias, which can show up in a context as fundamental as ordinary least-squares regression. This answer shows a simple example, in which adding a new categorical predictor actually flips the direction of a continuous ...


1

This is the purpose of the sklearn pipeline. Jointly optimizing PCA components and SVM parameters can be done using model selection tools like cross-validation or similar. The User Guide has more information. https://scikit-learn.org/stable/modules/compose.html#pipeline


1

The difficulty with small sample size and data-driven model optimization (such as recursive feature elimination) is that the optimization needs sufficiently certain estimates of the target function (importance). While (in)stability and thus uncertainty depends on the figure of merit, in general, the fewer cases you have to calculate the figure of merit, the ...


1

It sounds like you have a response that has a very non-linear relationship to the features you're using. By itself, X may have a weak relationship to the target, but in combination with another feature, it becomes important. If you're using neural nets or random forest (or other decision tree ensembles) that can identify interactions between variables, ...


Only top voted, non community-wiki answers of a minimum length are eligible