Improved Linear Regression Forecast Algorithm Based On Association Rules

Data Mining is a kind of process that reveals potential useful knowledge from massive data. In recent years, data mining technology has become the core technology in the field of commercial sales,and are widely applied to many fields,it causing a great concern in the academia and industry circle,so how to improve the efficiency of data mining becomes a hot topic in the academic field. This article is mainly aimed at association rules founding, commercial sales practices in the field of applied research.This article discusses the general theory of data mining, including the concept of data mining,tas,model,applications and trends of data mining. Studied the classical Apriori algorithm about association rules. and study of the linear regression analysis prediction algorithms, Summed up the problems of the linear regression analysis prediction algorithms,put up an improved linear regression analysis prediction algorithm based on association rule. The main idea of the algorithm is to first use linear regression analysis algorithms to predict the sales of a drugs, then use the improved linear regression analysis algorithm to predict the sales of associated with each other drugs. Compared this algorithm with linear regression analysis algorithm, it reduces the time costs, improve the efficiency. Meanwhile, I have designed a linear regression prediction system according to Apriori algorithm and improved linear regression analysis algorithms, The system's main function is to mine frequent itemsets, From the frequent itemsets (not including frequent 1-itemsets),selected an itemset of the largest number of support, using an improved linear regression analysis algorithms predict its's sales in the future of few months. Finally, The linear regression prediction system applied sales of drugs,to forecast sales of drugs, And predicted results are verified. Showed an improved linear regression prediction algorithm is feasible.The certain results has been obtained from this paper, but there are still many questions to be further studied, for example, to improve the accuracy of prediction; and how to present the mining results friendly to the user.