Thanks to IoT technology, industries are becoming more connectedleading to the emergence of Industrial IoT technology, the so-calledIIoT. This capability of being connected through the internet has cre-ated several security breaches that can be used by adversaries to getaccess to secret information or to corrupt the correct functionalityof the system. Consequently, security has become a serious concernin the design of such networks. Although a huge amount of workand research has been done before, most of them are computation-ally dependent on the cloud which is not completely suitable for real-time demand. On the other hand, the limited resource setting of theedge device does not allow us to implement complex algorithms. Inthis thesis, I intend to propose a lightweight machine learning-basedanomaly detection approach (specifically Autoencoder) on the edgeof the network to efficiently detect abnormal data patterns in networktraffic caused by different security attacks.