Long Range Wide Area Network (LoRaWAN) introduces the Adaptive Data rate (ADR) mechanism [1] aiming to maximize both battery life of the end-devices and overall network capacity. ADR performs adaptive tuning of radio configurations of the nodes by adjusting bandwidth, spreading factor, coding rate and transmission power parameters whenever the signal quality changes. The ADR algorithm was established for stable radio channel environments and is not efficient when conditions dramatically change (e.g. mobility). So, we have previously proposed an Enhanced-ADR (E-ADR) [2] that deals mobile nodes in case of predefined mobility patterns. However, several Internet Of Thing (IoT) applications, such as smart cattle ranching in smart farms [3], require sensors travelling with unknown or undefined trajectories. So, this paper extends E-ADR to unknown mobility pattern. This E-ADR extension, called VHMM-based E-ADR, is based on a Variable order Hidden Markov Model (VHMM) to predict the node trajectory. It has been implemented on Waspmote SX1272 hardware platform. Experimental results show its high efficiency in terms of the packet loss rate (PLR) and the energy consumption.