Predicting Itemset Sales Profiles with Share Measures and Repeat-Buying Theory

Given a random sample of sales transaction records (i.e., scanner panels) for a particular period (such as a week, month, quarter, etc.), we analyze the scanner panels to determine approximations for the penetration and purchase frequency distribution of frequently purchased items and itemsets. If the purchase frequency distribution for an item or itemset in the current period can be modeled by the negative binomial distribution, then the parameters of the model are used to predict sales profiles for the next period. We present representative experimental results based upon synthetic data.