Numerical wave models along with satellite and field observations are sometimes unsuccessful in giving accurate forecasts during extreme cyclonic periods and thus come the importance of statistical measures and during recent time artificial intelligence methods. In this study, statistical parameters like wave spectral skewness which measures wave asymmetry and wave spectral kurtosis which measures wave steepness are investigated for tropical cyclones of different categories occurring in the Bay of Bengal. Skewness and kurtosis describe the shape and pattern of the ocean waves. Deep learning models are introduced to estimate skewness and kurtosis parameters during cyclonic times. The transition month of May is chosen which witnessed cyclones of various intensities in the basin of Bay of Bengal, and for this study, the super cyclonic storm AMPHAN (May 2020), the extremely severe cyclonic storm MOCHA (May 2023), the very severe cyclonic storm YAAS (May 2021), the severe cyclonic storm ASANI (May 2022), and the cyclonic storm ROANU (May 2016) are considered category wise. For the storms of various categories considered, there is an increased value (up to 6 times) of the kurtosis or wave steepness with the increase of cyclone intensity although the skewness does not change in magnitude significantly with increased cyclone intensity. Training the skewness and kurtosis data using the long short-term memory (LSTM) model which is a deep learning model shows higher root mean square error (RMSE) values (50% approx) for intense cyclone kurtosis data. To study wave variability and pattern during extreme events, such deep learning-based predictions are essential as they pave way to low cost accurate forecast and hence better hazard mitigation.