Prenatal depression is a mood disorder that affects women during pregnancy period. If left undetected or untreated, can lead to serious consequences for both the mother and the child. Traditional detection methods primarily rely on self-reported questionnaires, which often suffer from subjectivity and limited diagnostic accuracy. To address these challenges, this paper proposes a Hybrid Deep Learning architectures for predicting prenatal depression using multimodal depression assessment data. The proposed architectures include CBLL which integrates a Bidirectional Long Short-Term Memory (BiLSTM) layer with two LSTM layers and CBGG which combines a BiLSTM layer with two Gated Recurrent Unit (GRU) layers. Both models are implemented with Weighted Average Pooling technique with assigned odd and even weight sequences to enhance feature representation and learning efficiency. The models were evaluated using performance metrics, including accuracy, sensitivity, specificity and AUROC. Experimental results demonstrate that the CBGG model outperforms the CBLL model, achieving superior accuracy. Furthermore, when comparing the weighted pooling strategies, the even weight sequence of the CBGG model achieves the best overall performance. These findings highlight the potential of Hybrid Deep Learning approaches for accurate, efficient for detection of prenatal depression.