The ongoing digitalization and the resulting networking of a heterogeneous landscape of devices require the development of novel security concepts and the creation of innovative mechanisms to protect them. This work addresses this problem and presents an Artificial Intelligence optimized heuristic network analysis model. The system provides a generalizable and flexible upstream mechanism for data collection, mapping, prioritization, and processing for classification of network traffic in the context of network intrusion detection systems. The processing of network packets regardless of their cryptographic state is enabled by metadata based feature extraction. A queue based aggregation of network packets in three layers makes it possible to create a novel data space that combines architectural and topology characteristics of the model structure with time series calculated features. This novel feature space integrates condition information of the model from various states within a time interval and is used to incorporate a prioritization mechanism for the optimized processing of network traffic. The features developed with a focus on Artificial Intelligence reflect the model architecture and states and serve as the basis for anomaly and intrusion detection using machine learning methods. Isolation Forest anomaly detection models to identify anomalies and Random Forest and Light Gradient-Boosting Machine classification models to binary classify attacks and to classify specific attack vectors are developed and evaluated. The developed demonstrator integrates the model logic and enables the processing of real-time captured network data as well as recorded intrusion detection data sets. The evaluation is carried out using the data sets TON IoT and CIC IDS2017 as model input. The time interval features of the sub-data spaces serve as the training basis for the development of the machine learning methods. The results show that the developed model architecture and topology features, computed over fixed time intervals, provide results ranging up to very good machine learning detection performances, depending on the respective threat vector and attack technique. Evaluation of Machine Learning models also shows that network metadata information represented in the structure of the model functions as flexible and generalizable feature vector space, which can be used effectively to achieve synergistic prioritization effects.