Mining multivariate time series for product migration analysis

As new technologies or products emerge, customer may migrate from a legacy product to a new product. One way to find out who migrate, how migrations look like, and the relationship between the legacy product and the new product is through mining the customer transaction history over time. For these purposes, we propose two customer segmentation procedures to quantify business impact of technology substitution. By assuming a general linear relationship between two substitutable products, we first develop a co-integration model to describe the dynamic relationship of two substitutable products. We then add structure breaks in the co-integration model to capture business changes along time. Structure breaks in either slope or intercept of the linear model are considered and the Least Angle Regression (LARS) algorithm is applied to estimate the structure break co-integration model. The estimated parameters are used to segment migration customers. The contribution of the proposed method is the mining of multivariate time series relationships of various customers, which is different from traditional time series mining research where univariate similarities among time series of different customers are explored. Another advantage is the inclusion of multiple break points with different types in a unified co-integration framework. To validate the accuracy and efficiency of the model, an industrial example from a telecommunication company is demonstrated.