The invention relates to the field of arrearage, power failure and power restoration complaint tendency prediction, in particular to a random-forest-method-based arrearage power failure and recovery complaint tendency prediction method. The complaint tendency prediction method comprises the following steps: obtaining a source data width table which comprises complaint behavior information of a user caused by arrearage stop and power restoration, wherein the source data width table comprises N final indexes; generating I decision-making trees by the source data width table, giving a result about whether a user has complaint tendency by each decision-making tree, and forming a random forest model by the I decision-making trees; inputting the source data width table of the user into a randomforest model, judging whether the user has a complaint tendency or not through I decision trees in the random forest model, and if the decision tree exceeding a set threshold judges that the user hasthe complaint tendency, voting the user as the user with the complaint tendency by the random forest model; Otherwise, the random forest model votes the user as a user without complaint tendency. Theprediction method can improve the accuracy of predicting the customer complaint tendency.