Severe typhoon disaster may lead to power outage of distribution network users, the effective prediction of the number of outage users in distribution network can provide auxiliary guidance for emergency repair of power grid. Considering meteorological factors, power grid factors and geographical factors, a forecasting method for the outage number of distribution network user based on machine learning regression algorithm is proposed. This paper analyzes and compares the application effects of five machine learning regression algorithms, including Linear Regression (LR), Support Vector Regression (SVR), Classification and Regression Tree (CART), Gradient Boosting Decision Tree (GBDT) and Random Forest (RF) in forecasting the number of power outages of distribution network users. The results show that LR performs poorly in forecasting the number of power outages of distribution network users, followed by SVR and CART, RF and GBDT are relatively good. The error of GBDT is close to RF, but considering that GBDT is serial computing and RF is parallel computing, RF is more efficient when used. Therefore, RF is selected for further analysis of the forecasting effect of outage number. The results show that the accuracy of the method is more than 70% when the error is within ±30%. It can provide powerful guidance for power outage repairs for distribution network users. This work is supported by the Science and Technology Project of China Southern Power Grid (No. GDKJXM20198441) and the Cooperative Education of Industry-academy Cooperation of Ministry of Education (No. 201902056044).