Due to the increasingly fierce competition in the electricity market, requirements of clients for service quality are constantly improving, and the number of complaints of clients continues to rise. Based on the architecture of the power customer complaint prediction model based on big data, a power customer complaint prediction method based on local linear embedding and deep forest algorithm is proposed in this paper. The local linear embedding algorithm is used to reduce the dimensionality of the input feature vectors of the customer complaint prediction model to reduce the computation and avoid flling into the local optimum solution. The dimensionality reduced feature vector of complaint prediction is scanned with multi-granularity to improve its representational learning ability. The deep forest algorithm model is established based on the cascade forest to realize the complaint prediction of customers. The simulation results of actual data show that ,compared with no dimensionality reduction processing and other prediction models, the prediction model proposed in this paper can more accurately predict the trend of customer complaints, which provides a reference for the analysis and prediction of customer complaints by power companies.