Abstract: Traditional frequent power outage management mainly uses manual multi-system queries and manual calculations for statistical analysis, resulting in heavy workloads and incomplete data analysis, which severely restricts the scientificity, advancement, and lean level of distribution network management. This paper structures and standardizes fundamental data, frequent power outage data, and line-transformer-customer data related to power outages, forming big data ecological collection and management to mine ‘digital value’. It uses both Support Vector Machines (SVM) and Logistic Regression for simultaneous prediction to reduce classification error probability, with improvements made through Random Forest algorithm. From the perspective of customers’ actual electricity usage experience, new classification models and calculation methods for customers’ power supply sensitivity level are studied. A combined approach using multiple machine learning methods is adopted to predict customer complaint probability based on customer sensitivity and relevant characteristic data of power outage events.