Complex enterprise environments consist of globally distributed infrastructure with a variety of applications and a large number of activities occurring on a daily basis. This increases the attack surface and narrows the view of ongoing intrinsic dynamics. Thus, many malicious activities can persist under the radar of conventional detection mechanisms long enough to achieve critical mass for full-fledged cyber attacks. Many of the typical detection approaches are signature-based and thus are expected to fail in the face of zero-day attacks. In this paper, we present the building-blocks for developing a Malicious Activity Detection System (MADS). MADS employs predictive modeling techniques for the detection of malicious activities. Unlike traditional detection mechanisms, MADS includes the detection of both network-based intrusions and malicious user behaviors. The system utilizes a simulator to produce holistic replication of activities, including both benign and malicious, flowing within a given complex IT environment. We validate the performance and accuracy of the simulator through a case study of a Fortune 500 company where we compare the results of the simulated infrastructure against the physical one in terms of resource consumption (i.e., CPU utilization), the number of concurrent users, and response times. In addition to an evaluation of the detection algorithms with varying hyper-parameters and comparing the results.