Tropical Cyclone Intensity Prediction using Bayesian Machine Learning with Marine Predators Algorithm on Satellite Cloud Imagery

Due to its wide range of associated hazards, tropical cyclones (TC) become the costliest natural disaster worldwide. A correct diagnosis model for the TC intensity can save property and lives. Unfortunately, intensity forecasting of TC has been a bottleneck and has made it difficult to forecast weather. Several existing approaches and techniques make a diagnosis of TC wind speed through the satellite data at the specified time with varying success. Deep learning (DL)-based intensity forecasting has recently held great promise in surpassing conventional approaches. DL-based techniques have been developed in geosciences to replace traditional methods. However, weather forecasting is uncertain due to the Earth system's nonlinearity, complexity, and chaotic effects. Thus, this manuscript develops a new Bayesian Machine Learning with Marine Predators Algorithm for TC Intensity Prediction (BMLMPA-TCIP) approach. The major goal of the BMLMPA-TCIP model is to estimate the level of the TCs on satellite cloud images. To accomplish this, the BMLMPA-TCIP technique utilizes the Gaussian filtering (GF) approach to eradicate the noise in the cloud images. Additionally, the extraction of useful feature vectors is performed by using the capsule network (CapsNet) technique. Moreover, the MPA method accomplishes the hyperparameter tuning of the CapsNet method. Lastly, the BMLMPA-TCIP technique utilizes the Bayesian Belief Network (BBN) method to predict TC intensity. To authorize the performance of the BMLMPATCIP approach, a wide variety of experiments are performed under the TC image dataset. The experimental validation of the BMLMPA-TCIP approach illustrates a superior RMSE value of 5.89 over existing techniques.