An Intrusion detection system is designed to classify the system activities into normal and abnormal. We use a combination of machine learning approaches as to detect the system attacks. The experimental results of the study show that increasing the number of classifiers has a threshold limit and the system accuracy will remain constant if the number of classifiers goes beyond this limit. The determination of the threshold limit is tentative. This article, also, presents a solution for unbalanced data of some attacks. The comparison of the results with other similar articles proves the efficiency of the presented system.