Parameter Reduction Optimisation for Analysis of E-Commerce Consumer Purchase Patterns

This research seeks to use parameter reduction optimization to improve the understanding of consumer behavior gained from Association Rule mining where the density of the data set is low. Low density occurs where small basket sizes are the norm, where very few and frequently only single items are purchased in each transaction and when there are large numbers of items being tracked across large numbers of transactions. The objective is to understand and optimize the aggregation levels that will provide consumer behavior insights of sufficient quality to the holders of relevant data sets. To achieve this goal, we analyze a publicly available dataset using a time-series Association Rule analysis for multiple time-period baskets contained in the dataset and at multiple item categorization levels with a view to optimizing the support levels of the output association rules. Purchases per consumer are viewed across days, weeks months and quarters to determine more suitable periods for analysis. The strength of the relationships is increased by also analyzing the data at the item category level where all items of a given category are correlated. We observe that increases to the support of the generated rules sets can be achieved, and that increase is dependent on the level at which the analysis is performed.