If you perform feature selection on all of the data and then cross-validate, then the test data in each fold of the cross-validation procedure was also used to choose the features and this is what biases the performance analysis.
Consider this example. We generate some target data by flipping a coin 10 times and recording whether it comes down as heads or tails. Next, we generate 20 features by flipping the coin 10 times for each feature and write down what we get. We then perform feature selection by picking the feature that matches the target data as closely as possible and use that as our prediction. If we then cross-validate, we will get an expected error rate slightly lower than 0.5. This is because we have chosen the feature on the basis of a correlation over both the training set and the test set in every fold of the cross-validation procedure. However, the true error rate is going to be 0.5 as the target data is simply random. If you perform feature selection independently within each fold of the cross-validation, the expected value of the error rate is 0.5 (which is correct).
The key idea is that cross-validation is a way of estimating the generalization performance of a process for building a model, so you need to repeat the whole process in each fold. Otherwise, you will end up with a biased estimate, or an under-estimate of the variance of the estimate (or both).
HTH
Here is some MATLAB code that performs a Monte-Carlo simulation of this setup, with 56 features and 259 cases, to match your example, the output it gives is:
Biased estimator: erate = 0.429210 (0.397683 - 0.451737)
Unbiased estimator: erate = 0.499689 (0.397683 - 0.590734)
The biased estimator is the one where feature selection is performed prior to cross-validation, the unbiased estimator is the one where feature selection is performed independently in each fold of the cross-validation. This suggests that the bias can be quite severe in this case, depending on the nature of the learning task.
NF = 56;
NC = 259;
NFOLD = 10;
NMC = 1e+4;
% perform Monte-Carlo simulation of biased estimator
erate = zeros(NMC,1);
for i=1:NMC
y = randn(NC,1) >= 0;
x = randn(NC,NF) >= 0;
% perform feature selection
err = mean(repmat(y,1,NF) ~= x);
[err,idx] = min(err);
% perform cross-validation
partition = mod(1:NC, NFOLD)+1;
y_xval = zeros(size(y));
for j=1:NFOLD
y_xval(partition==j) = x(partition==j,idx(1));
end
erate(i) = mean(y_xval ~= y);
plot(erate);
drawnow;
end
erate = sort(erate);
fprintf(1, ' Biased estimator: erate = %f (%f - %f)\n', mean(erate), erate(ceil(0.025*end)), erate(floor(0.975*end)));
% perform Monte-Carlo simulation of unbiased estimator
erate = zeros(NMC,1);
for i=1:NMC
y = randn(NC,1) >= 0;
x = randn(NC,NF) >= 0;
% perform cross-validation
partition = mod(1:NC, NFOLD)+1;
y_xval = zeros(size(y));
for j=1:NFOLD
% perform feature selection
err = mean(repmat(y(partition~=j),1,NF) ~= x(partition~=j,:));
[err,idx] = min(err);
y_xval(partition==j) = x(partition==j,idx(1));
end
erate(i) = mean(y_xval ~= y);
plot(erate);
drawnow;
end
erate = sort(erate);
fprintf(1, 'Unbiased estimator: erate = %f (%f - %f)\n', mean(erate), erate(ceil(0.025*end)), erate(floor(0.975*end)));
This is a general problem of selecting Hyper parameters of a model using CV. Feature selection is just a private example.
The same question arises with regularization coef selection, tree depth in selection tress, amount of layers in a NN of any other Hyper parameter.
Basically, the problem is that you can't use the same CV procedure to do both - select your Hyper parameter, and test your performance on the selected parameter.
Since you are overfitting in the CV space.
The best solution is to perform nested CV, where the inner CV simulates the process of Hyper parameter selection on each train fold, while the outer CV tests the performance of the model (after the inner param selection) on each test fold.
This process will provide a good generalized CV score, regardless of which values of the Hyper parameter (or which features in your case) where selected in each train fold internally.
How do you eventually chose the features you will use? now that you have a proper non-overfitted estimation of the performance of the model, you perform the same selection process on the entire dataset.
Just run the same feature selection process you used in each inner CV (on each train fold) on the entire dataset, and you have your final feature selection.
This is a known topic, you can find more here (and in a lot of other references) : nested CV answer
Best Answer
I see two possibilities:
You could conceivably use the MAPE directly as an objective function. Zheng (2011), International Journal of Machine Learning and Cybernetics) could perhaps be adapted to yield a smooth approximation to the MAPE, so you get a nice gradient.
You could try to get full predictive densities from your model and then derive the one-number summary of that distribution that minimizes the expected MAPE.
Yes, both of these do not address feature selection. You may want to take a look at this for a few ideas on this aspect.