On a recent project because of the characteristics of the data is not clear, but using machine learning classifying data, but encountered a problem,
Training sample 3000 or more, each data 3200 features, that is, the training sample is 3000 x3200 matrix, PCA dimension reduction to 3000 x512, training and test results are also can,
But the problem is in the practical engineering use, I need to a single data to estimate that this is caused by too little sample cannot be predicted, please the great god give directions should be how to deal with this situation?