In this paper, we propose an approachfor improvement of detction accuracy ofIntrusion Detection System (IDS). The traditional IDS based on number ofpackets detects intrusion when thenumber of target packets exceeded theprespecified threshold. However it isdifficult to set the appropriate valueas the threshold because the conditionof network traffic changed from hour tohour. Therefore, in order to improvedetection accuracy, we use a k-NearestNeighbor approach instead of thethreshold-based approach.