In multiple input multiple output orthogonal frequency division multiplexing (MIMO-OFDM), an efficient pilot pattern design (PPD) and channel state estimation (CSE) are crucial to ensure the consistency and heftiness of pilot-based CSE methods. Numerous methods have been developed to address pilot contamination and CSE issues to mitigate dynamic environments with nonlinear channel properties. Nevertheless, these methods involve multiple channels at recipient side, which causes delay dispersion and communication intrusion. Hence, to address these issues, this study developed the deep learning (DL) algorithm to obtain channel state information (CSI) grounded on CSE. The hybrid convolutional neural network (hybrid-CNN) and rat swarm optimization (RSO) are proposed for optimal pilot designing and selecting the location of pilot. Further, long short-term memory (LSTM) method is introduced for CSE in the MIMO-OFDM system. The presented approach results are compared using several metrics, such as mean squared error (MSE) and bit error rate (BER). The proposed method attains a minimum BER of 0.0012 × 10^(−3) and an MSE of 0.01 for the SNR of 20 dB, it demonstrates that hybrid-CNN-LSTM outperforms other state-of-the-art methods like recurrent neural network (RNN) and gated recurrent unit (GRU).