This document discusses a multi-stage intrusion detection approach that integrates unsupervised, supervised, and outlier detection methods. It evaluates the proposed approach on several datasets and finds it performs better than current methods in terms of accuracy and detection of known and unknown attacks while reducing false alarms. Specifically, it combines unsupervised clustering, supervised anomaly detection using support vector machines (SVM), and outlier analysis using the k-point+ and GBBK+ methods in a staged framework. Simulations show the approach effectively manages outliers and missing data. A complexity analysis finds the proposed method and prior approaches have the same asymptotic complexity but the new method provides better results for large datasets. Performance analysis using SVM on datasets both with and without outliers identified by the