Basket Analysis

In the dissertation, we researched on the two questions in market basket analysis: most profitable item selection and profits prediction in price promotion. We proposed the ways to integrate data mining techniques into user object directly. Experiments show the ways are efficient and effective.We proposed MOPIS (MOst Profitable Items Selection) to select the most profitable items. MOPIS decides the best selection after predicting some selections, and decides the profits of a selection by predicting the transactions generated by the selection. MOPIS models customers' buying well, can predict the buying of other items when the wanted items are unavailable, and also can predict the effects from the unavailable items. MOPIS applies a new and efficient way to estimate the profits of a selection, builds a new framework for item selection, and proposes a new way to predict the count and probability of the substituted items by calculating the distances between items. We proposed a new way to predict the cross-selling effects between different items. We sort the cross-selling effects between items by the contributed profits. We proposed a heuristic algorithm and a GA based algorithm, gaMOPIS, to select the most profitable items.We proposed PEPP (Profits Estimation in Price Promotion) to predict the profits in price promotion. PEPP predicts the transactions in the promotion by analyzing the transactions before the promotion. PEPP models well customers' buying in price promotion, can predict the buying of other items when the prices are changed, and also can predict the effects from the promoted items. PEPP proposes a new way to predict the count and probability of the items in price promotion by calculating the distances between items, gives a new way to predict the effects from the promoteditems by using the cross-selling effects, and proposes a new way to predict the buying of the promoted items from the customers who didn't buy the promoted items before price promotion.From the research, we get the results: Both MOPIS and gaMOPIS are efficient algorithms to select the most profitable items; Experiments show that MOPIS can fast find the selection with high profits, and it is 14 times faster than MPIS and it is effective and efficient for data sets with large number of items and transactions; PEPP is an efficient algorithm to predict the profits in price promotion; Experiments show that PEPP can fast predict the profits effectively and it is also effective and efficient for the data set with large number of items and transactions.