As the cost of information processing and Internet accessibility falls, organizations are becoming gradually defenceless to potential cyber threats such as network intrusions. So, there exists a need to run secure and safe transactions through the use of Intrusion Detection Systems, authentication, firewall and other hardware and software solutions. The existing Intrusion Detection system abilities to be adapted are very limited. This makes them ineffective for new or unknown attacks detection or to be adapted to an evolutionary environment. Machine learning approaches offer a potential solution to adaptation and correctness problems in Intrusion detection.Some Intrusion Detection systems does not deal with real time high speed networks. The high false positive rate is another issue with existing intrusion detection systems. In this paper, we present the machine learning approach for Intrusion Detection system which helps to reduce the false positive rates and increase the classification accuracy. We are going to train our system using the Real time data set using Naïve Bayes machine learning algorithm. The role of our system is to attempt to trap an adversary's attendance on a compromised network. Our System notices vulnerable packets that are trying to come into the network. We capture live packets and extract only the relevant header features.This improves the accuracy of the proposed system.Finally, using Naïve Based off-line trainer, we were able to achieve 90.2233 percent accuracy using Cross Validation of 10-fold and 76.6812 percent using supplied test dataset while maintaining 0.102 false positive rates.