The conditional independence assumption of the naive Bayesian（NB）algorithm sacrifices the classification accuracy to some extent

and it is particularly prominent in dealing with complex network intrusion data.Aiming at this problem

a feture-weighted Jensen-Shannon divergence and reverse class frequency（RCF）improve naive Bayes（JRNB）intrusion detection algorithm is proposed.Firstly

in order to compensate for the deficiency of the equal analysis of all feature items in NB algorithm

the Jensen-Shannon（JS）divergence method is introduced to measure the weight of each feature item for highlighting the difference between different feature items.Then

considering the influence of the class frequency on the sample classification

the inverse class frequency（RCF）is proposed to improve the calculation of the feature weight

further reducing the impact of conditional independence.Simulation results show that the JRNB algorithm achieves better detection performance in the appli-cation of intrusion detection.