Research And Implementation Of Sales Forecast In Hadoop-based Enterprise Marketing System

With the development of the Internet in recent years and its application in enterprises, it not only accelerated the development of various emerging industries, but also has brought opportunities and challenges for the various enterprises. Enterprises have a large number of marketing data and customer information, the information accumulated and its mining provides businesses with effective marketing channels. With the passage of time, the company’s marketing data has become increasingly large, which has reached the level of TB. The hardware equipment and software which are initially built by the enterprise don’t have the ability to store and process the massive data, so that the valuable information can not be extracted. A model which built by these marketing data can be used to dig out the useful information of enterprise’s management, and then the enterprise can make marketing strategy based on the information. Sales forecast is based on the history of sales to forecast the future, and it is a tool for the enterprise to determine the amount of production, to ensure that goods supply and marketing balance, which can help companies make the right business decision making to maximize profits. This paper constructs business marketing system based on Hadoop, andrealizes the storage, cleaning and processing of massive data. Researching on sales forecasting model, the combination forecasting model is established based on the ARIMA model and the BP neural network model, which is applied to the enterprise marketing system, providing marketing strategy support. The detailed contents about research are as follows:1. In view of the enterprise marketing data it has a complex data structure, large scale, data correlation and so on. This paper put forward the construction of enterprise Hadoop data processing platform, designed and implemented the MapReduce computing model to the relational data parallel computing, and solved the problem of enterprise’s distributed storage and processing of massive data.2. Due to the diversity of enterprise marketing data, there are many missing, wrong, inconsistent and redundant data in massive data. This paper analyzed its causes and forms, designed a cleaning method of "dirty data" in the system, and improved the Filter Bloom algorithm, in addition, proposed and implemented based on the tree structure of the Filter Bloom algorithm for repeated data cleaning.3. In order to further improve the accuracy of sales forecast, this system integrated a number of sales analysis model, and put forward the combination forecasting model based on ARIMA and BP neural network model. Moreover, MAE, RMSE and MAPE forecast evaluation criteria were used to compare and analyze the forecast effect of the combination forecasting model and a single ARIMA prediction model and BP neural network model, for predicting the effect carries on the contrast analysis, and it was proved that the combination forecasting model is superior to the single model.4. According to the actual situation of enterprise marketing, combined with the sales forecast model of this paper, the enterprise marketing systembased on Hadoopwas designed, implemented and deployed. Therefore, enterprise sales analysis is no longer relying on SAP or ERP system, which has a professional marketing system for sales analysis, and provides effective information support for enterprise decision.