Cloud computing is the distribution of computing resources over the Internet. A shared pool of resources, including data storage space, computer processing power and applications are provided by Cloud computing. In spite of being attractive, it challenges with new security threats when it comes to deploying an Intrusion Detection System (IDS) in Cloud environment. It requires a lot of time to process the Cloud dataset and produce proper classification strategy. A Penalty-Reward based instance selection method to reduce the Cloud dataset is proposed here. Using this method all the noisy and boundary instances are removed from the training dataset. After that Reverse Nearest Neighbor Reduction (RNNR) method is applied on the remaining instances to select all relevant instances from them. This helps to reduce the training time as well as to produce better classification accuracy for IDS.