In recent years, the network infrastructure has been improved constantly and the information techniques have been applied broadly. Because the misuse detection and anomaly detection methods both have individual benefits and drawbacks, this paper supports the point that combines these two methods to construct the whole intrusion detection system by data mining technique. In this paper, we focus on the improvement of the anomaly detection module in MINDS(Minnesota Intrusion Detection System). By analysis, we use the method of multi-dimension outlier point detection and adapt the connection score with dynamic weight to improve the performance of intrusion detection system. The improved unsupervised anomaly detection algorithm, also named IUADA, is non-linear, and reduces both the response time and the false alarm rate.