Rainfall is an integral component of the natural environment and a crucial climatic factor for the survival of people, animals, and plants. Accurate prediction of rainfall is a very crucial task and must be reliable. In machine learning, input data plays an important role in producing a good predictive model. The primary single features cannot fully reflect the complex relationships and spatial-temporal ones hidden within the original data. In this work, we introduce a new methodology for feature engineering from primary features for rainfall forecast. To assess the effectiveness of the proposed method, various machine learning models are used and evaluated via different metrics such as Accuracy, F1-scores and F1-score weighted. According to the proposed technique, the predictive models create a remarkable improvement compared to the results obtained from original features in many different metrics.