The main objective of an expected maximization (EM) based envelop semi-supervised neural net learning approach in this paper is to develop a model for forecasting peak CPU usages under unpredictable web traffic (load conditions) in a large enterprise applications environment with several hundred live applications. This proposed approach forecasts the likelihood of extreme peak response time because of the stressed CPU due to a burst in incoming web traffic from the live IT applications and then predicts the CPU utilization under extreme load (peak) conditions. The enterprise complex IT infrastructure consists of many applications running simultaneously in real-time. Features are extracted after analyzing the CPU-load patterns of demand which are mainly hidden in the data related to key transactions of the IT applications. This method generates synthetic CPU load profiles by simulating virtual users and execute the key transactions in the test environment. This model is used to predict the excessive peak utilization under peaked CPU conditions. We have used envelope expectation maximization classifier method with forced learning, attempting to extract and analyze the parameters that maximize the likelihood of the model after marginalizing out the unknown labels. This has resulted in mitigating the risks of system failures and enabled us to manage the IT capacity within 3 days from 7 days. This model has helped in IT capacity planning and optimal usages of existing IT infrastructure with minimal risk.