An approach to evaluate time-dependent changes in feature constraints

Feature selections mining is the process of discovering potentially feature associations and constraints in data. Especially, mining from time-series data obtains feature constraint trends. In this paper, we describe an approach to evaluate feature constraint trends and present results of two case studies. Feature selections mining was applied to a product transactions database at Hitachi. The product transactions had 148 optional features, and 8,372 products were derived from the product line. Both case studies focus on transaction-time periods: time series and time intervals. Feature selections mining discovered feature constraints around 100 rules in each study, and determined they constantly change.