The optimal transmit power that maximizes energy efficiency (EE) in LoRa networks is investigated by using deep learning (DL) approach. Particularly, the proposed artificial neural networks (ANNs) is trained two times; in the first phase, the ANNs is trained by the model-based data which are generated from the simplified system model while in the second phase, the pre-trained ANNs is retrained by the practical data. Numerical results show that the proposed approach outperforms the conventional one which directly trains with the practical data. Moreover, the performance of the proposed ANNs under both partial and full optimum architecture are studied. The results depict that the gap between these architecture is negligible. Finally, our findings also illustrate that instead of fully retrained the ANNs in the second training phase, freezing some layers are also feasible since it does not significantly decrease the performance of the ANNs.