Industrial internet of things and operational technology (IIoT/OT) lead the edge use case implementations. 5G and multi-access edge computing (MEC) offer the means to implement IIoT scenarios, ensuring business growth and deployment protection against network attacks. A variation of MEC and IIoT security measures are studied in the literature, and intrusion detection solutions are consequently proposed - including machine learning based solutions for anomaly detection. Automated machine learning (autoML) frameworks aim to create high accuracy models for users with little expertise in machine learning. This paper suggests autoencoders to improve autoML best model performance on a learning task: binary classification of network traffic. The experiment was performed on a benchmark dataset with intrusion detection examples: Network Security Laboratory - Knowledge Discovery in Databases (NSL-KDD). In order to optimize the learning process, autoencoders are suggested for feature encoding. The approach presented in this paper achieves a 4% increase in model accuracy and lower training time, when compared to the AutoML baseline model.