Bidirectional gated recurrent unit (BGRU) can learn hierarchical feature representations from both past and future information to perform multi-class classification. However, its classification performance largely depends on the choice of model hyperparameters. In this paper, we propose a methodology to select optimal BGRU hyperparameters for efficient botnet detection in smart homes. A deep BGRU multi-class classifier is developed based on the selected optimal hyperparameters, namely, rectified linear unit (ReLU) activation function, 20 epochs, 4 hidden layers, 200 hidden units, and Adam optimizer. The classifier is trained and validated with a batch size of 512 to achieve the right balance between performance and training time. Deep BGRU outperforms the state-of-the-art methods with true positive rate (TPR), false positive rate (FPR), and Matthews coefficient correlation (MCC) of 99.28 ± 1.57%, 0.00 ± 0.00%, and 99.82 ± 0.40%. The results show that the proposed methodology will help to develop an efficient network intrusion detection system for IoT-enabled smart home networks with high botnet attack detection accuracy as well as a low false alarm rate.