Many industries rely on weather forecasting, including agriculture, transportation, and disaster management. Traditional weather forecasting models rely on complex numerical calculations and empirical approaches, which typically produce low accuracy and reliability. Deep learning developments, on the other hand, have opened up new opportunities for improving weather forecast accuracy. This paper looks at the use of deep learning algorithms in weather forecasting and presents a unique approach to enhancing prediction accuracy and reliability. Machine learning has achieved incredible success in a wide range of programmes, including predicting rain, a complex meteorological variable with significant societal ramifications. This study employs a novel form of gradient-boosted trees to forecast seasonal precipitation in the region of South America. To determine the appropriate hyperparameters to use with the gradient-boosting approach, the Optima framework employs Bayesian optimization. Seasonal precipitation predictions based on data collected from observation are juxtaposed with predictions produced by the Brazilian National Institute of Space Research (INPE, Portugal), the gradient-based boosting method, and deep neural network algorithms. The results reveal that the enhancing method outperforms other approaches in a variety of situations.