Is there any way to calculate P value in weka java?

Question

I am trying to develop a feature selection algorithm in java. I am using weka libraries for this purpose. Is there any way to calculate P value using weka libraries ? or is there any java machine learning library to calculate p value ?

I was able to calculate chi square value using weka libraries. Is there any way to calculate p value using this chi square value ?

Answer 1

You can use SignificanceAttributeEval , not chi-square. You can download it from the package manager (go to the Weka GUI chooser, tools > package manager ). This algorithm works only with a nominal class, but if you have a numeric class you can discretize in bins. This algorithm evaluates the worth of the attributes in relation to the class by computing probabilistic significance as a two-way function. Like the P-value, this algorithm measures the probability that two features happen together not by chance, but unlike the P-value, here the best features have a higher score. The score is similar to the one of correlations, but has only positive values, non-significative features have score 0.