I am struggling to find a way to perform better linear regression. I have been using the Moore-Penrose pseudoinverse and QR decomposition with JAMA library, but the results are not satisfactory. Would ojAlgo be useful? I have been hitting accuracy limits that I know should not be there. The algorithm should be capable of reducing the impact of an input variable to zero. Perhaps this takes the form of iteratively reweighted least squares, but I do not know that algorithm and cannot find a library for it. The output should be a weight matrix or vector such that matrix multiplication of the input matrix by the weight matrix will yield a prediction matrix. My input matrix will almost always have more rows than columns. Thank you for your help.
I don't fully understand your question, but I've used Apache Commons Math to do linear regressions before.
If you want to use a more generic external tool for this, use Octave. I think it's more suitable for these kind of things. If not, take a look to:
Logistic Regression in Java Specifically: http://commons.apache.org/math/userguide/overview.html
来源：https://stackoverflow.com/questions/8406305/is-there-a-java-library-for-better-linear-regression-e-g-iteratively-reweigh