ABSTRACT
CONTENTS 16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 16.2 Data Mining Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 16.3 Application to a Textile Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288 16.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
All along the items design process, the identification of parameters influencing the future sales remains one of the priority of the marketing director. In the apparel industry, textile items collections are mainly processed at meeting periods between the different partners of the firm. Each of them give their own impressions and subjective parameters to forecast the trends to come and the items attributes as well. In most cases, poor statistical tools are used to identify the relations between the items attributes and the historic sales. To fill this gap, we propose in this chapter a data mining technique aiming at explaining the historic sales profiles of items from the knowledge of their attributes. The first step of our approach is based on the PLS regression which allows us to extract the hidden relations (represented by the PLS factors) between the items attributes and their sales profiles. Based on these results, a particular hierarchical clustering of the sales season weeks can then be performed and constitutes the second step of our approach. More precisely, each week is identified by a set of sold items. If this items set remains almost unchanged for two consecutive weeks, then these two weeks are merged. As a consequence, the highest nodes of the hierarchy correspond to the main breakpoints of the sale season. Each of them put on the fore the main changes of purchase behaviors. Moreover, for each of these nodes, graphic plane representations can be obtained displaying weeks, items and items characteristics as well. Therefore, the comparison of the two groups of weeks defining a node and the interpretation of the associated breakpoint can be easily done by the inspection of its corresponding graphic representation.
