NONLINEAR STRATEGY TO CLASSIFY TIME SERIES OF THE SEMICONDUCTOR MARKET TREND

Abstract Semiconductor firm sales are a complex function of what direct customers require and how efficiently the company is able to satisfy them. The task of solving the management of production planning with a wide products portofolio is not an easy one. The paper deals with the adoption of an unsupervised clustering strategy to classify products, not just in relation with the business parameters, but also by considering the historical evolution of sales and customer demand. Different nonlinear techniques have been considered to face the problem both from the mathematical point of view than from the economic one. Particularly, this strategy has been implemented to classify the products of Discrete and Standard Ics Group of the worldwide semiconductor firm, STMicroelectronics in relation to their sales versus the different customer type. Two different approaches have been used: the hierarchical clustering with an optimization procedure and GHSOM structures. The results show the validity of the cluster strategy, in fact the time series carrying on the same qualitative information are grouped, and the flexibility of the GHSOM. Moreover, the GHSOM, as a clustering technique, requires a low supervision level and brings a reduction of the problem complexity and speeds up the process.