A Study Of Retail Company Commodities Sales Forecasts Based On Data Mining

With the expansion of economic globalization and the deadline of protection for our country from the WTO, the foreign-invested retail enterprises will unrestrictedly pour into China, facing those enormous competitions and challenges, how to survive is the most important issue for Chinese retail enterprises have to face with. It is the key to success that making sales forecasts and making the correct marketing decisions on the basis of this.In the commodities sales forecasts, the general commodities in the future for a certain period of the sales state and sales volume, the variables involved of forecast only the sale volume of commodities, a small part of taking into account the amount of sales, but has not been the most important factor, that is, profit, profit is the key factor of retail enterprises winning profit and development. Therefore, in this thesis, the sales forecasts for the selection of input variables, in addition to the commodities attribute used in sales volume of commodities and the amount of sales, seasonal index, etc, it also has been selected the important variable profit rates, for an increase of the accuracy of the forecasts.Using data mining technology in the thesis, I put forward a new commodities sales forecasting model——SPI-M model, which used in retail commodities sales forecasts help the enterprise operations and decisions-maker make important decisions. The procedure of the SPI-M model construction is: first , we analyse the marketing seasonal laws with the season analysis model (S model) based on statistics, calculate the seasonal index of commodities quarter and month. Second, we establish the profit rate grade model (P model) with the K-means algorithm on data mining technology, which Classifies the grade of profit rate .Third, we build decision trees model (I model) within seasonal index, profit rate grade, sales volume, the amount of sales as input variables of ID3 algorithm, obtain the sales state of yearly, quarterly and monthly historical data. Finally, we forecasted the transfer of sale state of commodity within a certain period in future with Markov model(M model) on statistics, its demand of market in the same period of time in future is the sum of the corresponding period average sales involve and average increasing sales involve in same time before.At last, the thesis established a commodity sale forecasting system based on SPI-M model. Meanwhile, we compare and analyse the results of SPI-M model, Season model and Markov model through experiments, and get the conclusion that we can reach higher correctness if we use SPI-M model than rest of forecast model.