Sales Forecasting Based On Combination Model

Under the background of big data assisted management decision,this paper establishes a time series sales forecasting model which is based on composite.The model is using random forest algorithm to purify the residual of the traditional time series model and obtain more accurate time series forecasting value.It will be helpful for support for business market planning,sales decision-making and operation strategy.This paper uses multi-dimensional and multi-sample data from Alibaba.And the paper designs prediction models based on ARMA and random forest respectively.Through a series of feature engineering such as data preprocessing,feature extraction and dimensionality reduction,a high-quality training data set is obtained.And through continuous optimization of parameters,a model with strong learning ability is built.Due to ARMA cannot extract the non-linear information well and the random forest model has strong learning ability for the non-linear information,combining the advantages of the two models,using random forest model to optimize ARIMA prediction residual.By this way,we got a prediction model with stronger prediction ability.At the same time,we got the influence of various factors on sales,which provides operational reference for store operators.