As an artificial intelligence method, machine learning (ML) has been widely used in prediction models of high-dimensional datasets. This study proposes an ML method, the Gradient Boosted Regression Tree (GBRT), to predict the intensity changes of tropical cyclones (TCs) in the Western North Pacific at 12-, 24-, 36-, 48-, 60-, and 72-h (hr) forecasting lead time and the model is optimized by the Bayesian Optimization algorithm. The model predictands are the TCs intensity changes at different forecasting lead times, obtained from the best track data of the Shanghai Typhoon Institute (STI) and the Joint Typhoon Warning Center (JTWC) from 2000 to 2019. The model predictors are the synoptic variables, climatological and persistent variables derived from the reanalysis data obtained from the National Centers for Environmental Prediction (NCEP), and the sea surface temperature (SST) data obtained from the National Oceanic and Atmospheric Administration (NOAA). The results show that the GBRT model can capture the TCs intensity changes well for the succeeding 12-h, 24-h, 36-h, and 72-h. Compared with the traditional multiple linear regression (MLR) model, the GBRT model has better performance in predicting TCs intensity changes. Compared with the MLR model, R2 of the GBRT model for TCs intensity forecast increases by an average of 8.47% and 4.45% for STI data and JTWC data. MAE (RMSE) drops by 26.24% (25.14%) and 10.51% (4.68%) for the two datasets, respectively. The potential future intensity change (POT), the intensity changes during the previous 12 h (Dvmax), Initial storm maximum wind speed (Vmax), SST, and the Sea-Land ratio are the most significant predictors for the GBRT model in predicting TCs intensity change over the Western North Pacific.