Rain can generally have convective and stratiform characteristics, associating with the different processes of cloud microphysics development which leads to differences in intensity, time and area of rain. Using dual-polarization radar observations in the Northwest region, the study applied the Support Vector Machine classification algorithm for classifying rain characteristics. The classification results are then used to find the empirical coefficients in the rainfall estimation formulas based on the Least Squares Method. The evaluation shows that with the proposed method, only less than 10% of the areas with bright bands are wrongly identified as convective rain, the method has also overcome some misclassifications compare to the reference method in some specific rain cases. Rainfall estimation results show that the estimated rainfall is lower than the observed values, the formulas with the classification of rain characteristics and using dual-polarization variables have better results than the formulas using only the of Z, in which the formula R (Z, ZDR, KDP) provides the highest correlation and the smallest error.