Hybrid beamforming has gained popularity in recent years due to its ability to enhance the performance of massive multiple-input multiple-output (massive MIMO) systems and improve the efficiency of wireless communication. However, designing a hybrid precoder is a challenging task as it requires accurate channel state information (CSI) and needs to solve an optimization problem. This paper introduces an unsupervised hybrid deep learning technique, namely CNN-BiLSTM, for developing hybrid beamforming in massive MIMO systems based on received signal strength indicator (RSSI) without having prior knowledge of CSI. Additionally, incremental principal component analysis (Incremental PCA) is used to reduce the dataset's dimensionality. The results of our proposed model demonstrate a significant improvement in spectral efficiency. This technique reduces the training overhead significantly and is able to quickly serve all the users. The simulation results indicate that proposed model produces low side lobes when providing coverage to areas. As a result, the performance of our model has a noticeable advantage compared to other models. In addition, proposed method achieves 99.21% accuracy and 5.30 bps/Hz sum-rate that is close to optimal performance and outperforms the state-of-the-art method.