Spam Filtering Issue: FPD Research between False Positive and False Negative
Zhen Liu, Zhou Ming-tian · 2007
According to the fact that False Positive is more serious than False Negative while doing spam filtering, novel Email filter with feature of partial dependency(FPD) is asked urgently. This paper investigates the FPD between False Positive and False Negative comprehensively and proposes an advanced fitted Logistic Regression model for spam discrimination by introducing a coefficient function involved with the feature of partial dependency. From four aspects including the precision ratio, dimensionality selection feature, KL divergence distribution between FP R and FN R , and noise withstanding, the new model is proved to be of evident CPD with respect to evaluation tests on real Email testing sets.