One realization of optical wireless communication (OWC) is optical camera communication (OCC), which utilizes mobile-phone and surveillance cameras for communication. One main limitation of OCC system is the low frame rate of typical cameras. To enhance the OCC data rate with limited frame rate cameras, rolling shuttered camera mode can be utilized. However, decoding the rolling shuttered pattern is very challenging. In addition, utilizing multi-level modulation formats, such as four-level pulse-amplitude-modulation (PAM4) can increase the OCC bit-per-symbol; however, it would be a great challenge to identify the multiple intensity levels owing to the uneven aggregated light exposure of each pixel-row. In this work, we put forward and demonstrate utilizing the Long-Short-Term Memory neural-network (LSTM-NN) to alleviate the decoding of high inter-symbol-interference (ISI) of PAM4 rolling shuttered pattern. The experiment is performed in a practical room with 3-m free-space transmission distance. Results show that the proposed LSTM-NN method can significantly enhance the decoding of multiple intensity level rolling shuttered pattern. The proposed scheme also allows a large tilting angle of ±70° between the transmitter (Tx) and receiver (Rx).