Virtualization technology plays a crucial role for reducing the cost in a cloud environment. Efficient virtual machine (VM) packing method that focuses on compaction of hosts such that most of its resources are used when it serves the user requests. Here our aim is to reduce the power requirements of a cloud system by focusing on minimizing the number of hosts. We propose a predictive scheduling approach considering the deadline of a task request and make flexible decisions to allocate the tasks to hosts. Experimental results show that the proposed approach can save around 5 to 10% power consumption than the standard VM packing methods in most scenarios. Even when the total power consumption requirements remain the same as that of standard methods in some scenarios, the average number of hosts required in the cloud environment are reduced and thereby reducing the cost.